Edtech software in Baltimore: half of City Schools’ tools waited on a vendor login, and AI skips Maryland’s accessibility check until 2028

Three people in Baltimore have a busy week ahead. The first is a former reading specialist in Remington who built a phonics app with a developer friend. A principal in Southwest Baltimore wants to try it with two second-grade classes, and the founders assumed a free pilot would need a handshake. It needs a memorandum of understanding approved by four offices, because City Schools requires a written agreement for anything that touches student data, and the district’s accessibility reviewer wants a test login and a conformance report the founders had never heard of. The second runs a two-person startup in Fells Point whose AI math tutor a Baltimore County middle school would like to try. Since June 1, Maryland law has required the district to score AI tools against a state rubric, and the rubric’s first question is whether the vendor contractually promises not to train its models on district data; the startup’s agreement with its model provider says nothing of the kind. The third owns a tutoring company in Towson that runs an after-school program for a City school under contract and sells printable practice tests to parents online. It keeps one email list for both, and Maryland’s student data law lets it market to parents only if the marketing did not come from what it learned through the school.

Each of them has a software problem dressed as a sales problem. The questions a Maryland school system asks before a student logs in, who sees which data, for how long, how it is deleted, whether a blind student can use the product, what the AI does with what a child types, are all answered by the way the product is built, and the answers have to exist as records before anyone asks. This article goes through those questions for Baltimore City and Baltimore County, using the statutes, the districts’ own procurement rules and reports, the state’s new AI rubric, federal rules, a census of 50 tools on Baltimore’s own lists and 80 small edtech companies nationwide, and what it costs to be ready. It also covers the other way to sell, directly to families, and why that store should be built as a separate one. It is our reading of the rules for the products we build, not legal advice.

Baltimore’s school market, by the numbers

On September 30, 2025, the Maryland State Department of Education counted 76,340 students in Baltimore City Public Schools and 107,983 in Baltimore County Public Schools: 184,323 children, or 20.9% of the state’s 880,231. Both systems are shrinking slowly, the City by 0.6% and the County by 1.9% in the last year, and the City by 10.2% since 2014. The City runs 150 schools, 30 of them charters; the County runs 177. These are among the largest buyers of education software within an hour’s drive of the Inner Harbor, and they are not the only ones: Anne Arundel, Howard, Harford and Carroll counties border them, and all 24 of Maryland’s school systems buy through the same state rules.

The money is large and the technology slice of it is thin. City Schools adopted a budget of $1.95 billion for fiscal 2027, 4.4% more than the year before; its Office of Information Technology gets $23.5 million of the general fund, 1.49% of it and 4.3% less than in fiscal 2026, with its contractual services cut by 12.6%. Baltimore County adopted $2.54 billion in all funds and $2.29 billion for operations; its information technology operations office gets $31.3 million, 9.7% less than last year, and the County’s budget book lists cuts of $3 million in device purchasing and leasing and $1.4 million in software license fees, noting that “digital learning devices are now provided for all students in all grades at every school.” In plain terms, the County already has a device for every student and both systems are now trimming their technology budgets, which makes a new subscription harder to add and a replacement easier to argue for.

Two sources of money that paid for a lot of edtech have ended. Federal pandemic relief, the Elementary and Secondary School Emergency Relief fund, gave Baltimore City $689.4 million and Baltimore County $337.4 million, 37.8% of everything Maryland’s school systems received, and it has been spent; the City’s budget book shows it at zero from fiscal 2026. When the Education Department abruptly ended late spending on March 28, 2025, the state put Maryland’s exposure at $232.1 million; Maryland joined a multistate lawsuit, and after two court orders the Department let states resume spending under their earlier approvals that June. And the state’s own Blueprint money for tutoring, called Transitional Supplemental Instruction, falls to zero in fiscal 2027 in both systems, from $8.9 million in the City in fiscal 2025 and $6.3 million in the County in fiscal 2024. Tutoring and intervention products that were bought with either fund are now competing for general-fund dollars.

The number of tools in use is high and getting harder to measure. Instructure’s annual EdTech Top 40 report counted an average of 2,739 distinct tools accessed per district in the 2023–2024 school year and 2,982 in 2024–2025, with each student using 48 tools a year. Its 2026 edition switched to counting launches from Canvas only and reported 3,001 tools available per district and four used per student or teacher, a break in the series rather than a collapse. In the same report, 52.5% of the top 40 tools had at least Tier IV evidence and 45% held a privacy certification. City Schools told its Board in July 2025 that nearly 90 of its platforms are reached through single sign-on and that it had no common library of them.

The companies that build this software are also local. The Census Bureau’s County Business Patterns for 2023, the latest year published, count 23 software publishers with paid employees in Baltimore City, employing 903 people at an average payroll of about $164,000, up from 489 employees in 2018, and 27 in Baltimore County employing 1,551, up from 693. Custom programming firms employ another 2,227 people in the City and 1,850 in the County. Educational support services, the Census category for educational testing, consulting and similar non-instructional services (tutoring is counted separately), employ 2,787 people at 44 establishments in the City, up from 1,552 in 2018, though the City figure carries the Census Bureau’s high-noise flag. The two jurisdictions hold 61.5% of Maryland’s employment in that category. Baltimore is also home to institutions any edtech founder will eventually meet, among them Johns Hopkins University’s Center for Talented Youth and its Center for Research and Reform in Education, which runs Evidence for ESSA, and the state adds TEDCO, the technology investment agency the General Assembly created in 1998.

One caveat about the money that founders should hear early: venture funding for education technology is still shrinking. HolonIQ counted $2.4 billion in global edtech venture funding in 2025 and $1 billion in the first half of 2026, 26% less than a year earlier. A Baltimore edtech company in 2026 is more likely to be funded by its first district contracts, or by families, than by a large venture round, which is one more reason to know exactly what the districts will ask for.

What it costs to be district-ready, and who pays the meters

Most of the gates between an edtech product and a classroom are free for the district and paid for by the vendor, and most of the vendor’s prices are not published. We looked for 34 things a small company might have to buy or join to sell to a school system, from rostering networks to security audits, on October 17, 2026. Sixteen have a published price, and five of those are zero; seventeen are “contact us” or say nothing at all; one pricing page refused to load from our location. A founder budgeting a first district contract is budgeting partly blind.

Getting students in the door

Rostering and sign-in come first, because no district wants teachers typing class lists into a startup’s product. Clever, which says more than 1,250 app partners roster through it, is free for schools, and its pricing page says plainly that “Application partners pay to roster their apps through Clever,” without saying how much; its terms bar a partner from passing that fee on to schools. Its free tier for app developers, Clever Library, has been discontinued, according to its developer documentation, although its pricing page still describes it. ClassLink takes the opposite position in writing: it “never charges companies to integrate.” Microsoft’s School Data Sync is a free service, and Google charges developers nothing to build Classroom add-ons, though those add-ons work only for teachers whose districts pay for Google’s Education Plus or Teaching and Learning licenses.

It also matters which systems Baltimore’s schools already run. City Schools uses Canvas as its learning management system, rolling out to high schools first, middle schools in 2026–2027 and elementary schools in 2027–2028; its student information system is Infinite Campus, and students sign in with Google accounts. Baltimore County uses Schoology as its learning management system and Focus as its student information system. A product that launches from inside Canvas and Schoology through LTI 1.3, and reads rosters through OneRoster, covers both. The standards body behind both specifications, 1EdTech, publishes no dues for any of its three membership tiers, certifies only members, and revokes every certification if the membership is not renewed. Its public directory listed 338 products with LTI 1.3 certification from 199 organizations and 64 with OneRoster 1.2 certification from 57 when we counted.

Badges, seals and the pledge that ended

Trust badges have changed more than most founders realize. The Student Privacy Pledge, the industry’s best-known promise, was retired by the Future of Privacy Forum on April 25, 2025, and its list of signatories stayed online only until July 31, 2025. Common Sense launched a Privacy Seal on April 3, 2025, which it says fewer than one in ten evaluated companies qualify for, and publishes no fee. ISTE’s Seal costs $5,000 for the review and $1,500 a year for the license. Digital Promise’s product certifications, including ones for ESSA Tier 3 and Tier 4 evidence and for responsibly designed AI, publish no current fee. The Global Education Security Standard from the consortium behind the national privacy agreement is free as a self-assessment, and a paid third-party audit announced on May 7, 2026 had no published price when we checked. The federal Cybersecurity and Infrastructure Security Agency’s Secure by Design pledge costs nothing.

Security and insurance

Security is where the real money goes. The cheapest published path to a SOC 2 Type I report we found was a $3,555 bundle from one assessor that combines it with the security assessment Google requires for apps using its most sensitive data scopes. Secureframe’s compliance platform starts at $7,500 a year before the auditor’s fee, and one consultancy’s September 2026 guide puts small-company audits at $5,000 to $10,000 for Type 1 and $30,000 to $70,000 for Type 2. Penetration tests start at $2,999 a year for an automated test of one application at Astra, $5,999 with certified testers, $3,500 per test at Cobalt under an offer that ends on December 31, 2026, and $10,800 at Software Secured. Insureon reports that its small-business customers pay a median $1,552 a year for cyber insurance, with premiums from about $400 to more than $8,000, but the limits districts ask for are not small-business limits: Montgomery County Public Schools’ June 2026 request for proposals for IT services required cyber insurance of at least $2 million per claim and in aggregate, and an Oregon district we checked requires $2 million of cyber coverage plus $2 million of technology errors and omissions coverage whenever student or staff data is exchanged.

Accessibility is one of the cheaper lines and, in Maryland, the most mandatory. Accessible.org charges $100 to $250 per primary screen for an audit, with most audits between $1,250 and $2,750, and then $350 to turn it into a conformance report in the WCAG edition of the template or $550 in the Section 508 edition. The edition matters here: §7–910 asks for a report on conformance to the Section 508 standards, so a Maryland bid needs the 508 or the combined international edition, not just the WCAG one.

None of the totals includes the prices nobody publishes, which are some of the important ones: what Clever charges a partner, what 1EdTech membership costs, and what an evidence study from LearnPlatform or a third-party security audit will cost. Nor do they include the engineering time to build the integrations, the deletion workflow and the accessibility fixes, which is usually the larger number. The typical column, about $26,400, is a large number for a three-person company, which is why we tell founders to buy these things in the order districts ask for them: the free integrations and the Section 508 report before the first bid, a penetration test before the first contract that holds student data, and SOC 2 when the second or third district asks for it rather than before the first one does.

How City Schools and Baltimore County buy software

The first thing a founder learns about selling to Baltimore’s two school systems is that the search results are not much help. The top result for how to become a vendor to Baltimore City Public Schools is a supplier page of the City of Baltimore’s own government, which is a different organization with a different procurement office, and a frequently surfaced Baltimore County Public Schools purchasing page now returns “not found.” The rules themselves are clear once you find them, and they start in state law.

The state threshold moved to $100,000 in 2024

Education Article §5–112 requires a county board to advertise for bids at least two weeks in advance when a purchase of supplies or equipment costs more than $100,000, in a newspaper, on eMaryland Marketplace Advantage (the state’s procurement site, known as eMMA) or on the school system’s own website, and to award the contract to the bidder that offers the best value, with the bidder’s plan for using minority contractors among the listed factors. Until July 1, 2024 the threshold was $25,000; Chapters 274 and 275 of 2024 raised it fourfold. Two exemptions in the same section matter to edtech. Contracts for “books and other materials of instruction,” as the state’s financial reporting manual defines them, are outside the bidding rule, and so is a board’s participation in a contract that another public agency or an intergovernmental purchasing organization bid publicly, which is the legal basis for most edtech piggybacking. A contract made in violation of the section is void.

Baltimore City Public Schools

City Schools buys in a decentralized way: each school or office chooses its own vendors within its own budget, while only the chief executive officer and the Director of Procurement may sign contracts, so a principal’s enthusiasm is not a signature. Under the Board’s Policy DJA and its administrative regulation, a purchase of $25,000 or less needs one quote, more than $25,000 up to $50,000 needs two, and more than $50,000 up to $100,000 needs three. A vendor without Board approval may not be paid more than $100,000 in a fiscal year across the whole district, and anything above that goes to the Board of School Commissioners, normally after an invitation for bids or a request for proposals. Each procurement item the Board sees must describe the solicitation, the cost, the expected outcomes and “evidence of effectiveness,” and federally funded purchases must meet the evidence rules of the Every Student Succeeds Act; the district keeps an evidence catalog for that purpose. Vendors register on eMMA to receive bid notices and on the district’s Oracle supplier portal to receive purchase orders and submit invoices, nothing may start before a purchase order exists, and payment terms are net 30 with no billing in advance.

One line in the regulation decides how a small edtech company should approach City Schools. For purchases of $100,000 or less a purchase order is normally enough, “except those involving digital tools and/or data sharing,” which need a written agreement. A free pilot is not a way around that: an arrangement in which City Schools pays nothing is handled as a memorandum of understanding, which needs the approval of the responsible chief officer, the district’s legal counsel, the Director of Procurement and the chief executive or the chief of staff. A product that touches student data enters City Schools through a contract or an MOU, either way with lawyers reading it, and the district’s published guidance on AI tells staff and students not to enter personal data into AI tools “without City Schools’ approved data privacy agreements.”

City Schools is also building the catalog that the 2022 privacy law imagined. On July 8, 2025 the Board approved a contract with Instructure for LearnPlatform, at about $150,000 a year through June 30, 2028. The Board item explains why: of the platforms staff and students use, “nearly ninety (90) are accessed through single sign-on,” but “a common library of tools is not available,” and the district wanted approved, pending and denied status flags, privacy flags, a request workflow for teachers and a view-only inventory for parents, partly to comply with the state’s accessibility law. We could not find a public City Schools library yet. When it appears, a vendor’s status in it will be visible to every principal and parent in the city.

Baltimore County Public Schools

Baltimore County’s Supplier Guide, revised in September 2026, sets an invitation for bids or a request for proposals at $100,000 or more, written quotes below $15,000 and at least two quotes from $15,000 to under $100,000, and registration in its Oracle supplier system with a W-9, good standing with the State Department of Assessments and Taxation and the commodity codes a vendor sells under. One of the district’s own purchasing pages still gives a $50,000 bid threshold, so read the guide. Instructional tools go through a curriculum committee under a superintendent’s rule before they are bought, and the district’s contract records for software carry a standard sentence: “A student data privacy evaluation was performed and the Student Data Privacy Requirements are included in the contract.”

The County publishes its contracts database, and it shows how much of the buying happens through other people’s procurements. Of 506 contracts in the August 2026 edition, 261 were awarded by invitation for bids, 61 by request for proposals, 78 through cooperative agreements and 42 through cooperative program administration, with 9 sole-source and 6 single-source awards. The cooperative contracts come from other school districts, the County government, national purchasing cooperatives such as TIPS, Sourcewell and OMNIA, and the Maryland Education Enterprise Consortium. The County’s student information system, Focus, was renewed as a sole-source contract of up to $4 million through 2030, and Microsoft licenses run through a consortium contract of up to $8 million.

The consortium, the neighbors and the free pilot

The Maryland Education Enterprise Consortium, hosted by the University System of Maryland since 1999, is the shortcut and the wall at the same time. All 24 Maryland public school systems are members, so a product on a MEEC contract can be bought by any of them without a separate bid, but a vendor can join only “by submitting a proposal to an open solicitation,” posted on eMMA, and new vendors are not added to existing contracts. In January 2025 it reported more than 200 member institutions and 98 vendors on 16 agreements; in August 2026 it issued a request for proposals for accessibility products and services. For a startup, the practical routes are a subcontract under a reseller that already holds a cooperative contract, a direct small purchase under $100,000, or a bid.

The neighbors show what a public catalog looks like. Anne Arundel County Public Schools publishes its LearnPlatform library, and on October 17, 2026 it listed 291 tools: 289 approved for use and 2 not approved, with 286 marked compliant on privacy, 3 unknown and 2 not applicable. Montgomery County vets every tool centrally for instruction, privacy, security and accessibility and publishes the results as a dashboard. And the most useful pilot story we found is from Baltimore County: TalkingPoints, a two-way translated messaging app for teachers and families, ran a districtwide pilot from January 1 to May 31, 2024, went through the curriculum committee in May, and was approved by the Board in June 2024 under a Fairfax County Public Schools contract, with spending authority that now stands at $1.1 million. The Board item cites the pilot’s findings: usage across the district and more engagement from staff and from families who received messages in their own languages. That is the path in miniature: a pilot with results the district can measure, a committee review, and a contract another district had already competed.

Maryland’s student data law, read as a data model

Maryland’s Student Data Privacy Act is Education Article §4–131, enacted in 2015 and rewritten in 2022, and it is short enough to read in one sitting. We read it the way we read any statute that a client’s software has to obey: as a description of the fields a database must hold, the actions it must be able to take and the actions it must be unable to take. Our article on tutoring companies walks through the basics; here we go further into what it asks of a product that a district actually buys.

When the law starts to apply

The Act regulates an “operator,” and the definition has three conditions that all have to be true. An operator is an individual or an entity that “engages with institutions under the school official exception” of the federal Family Educational Rights and Privacy Act, that operates “in accordance with a contract or an agreement with a public school or local school system in the State” to provide a website, an online service or an app, and whose product processes covered information and either is used for a PreK–12 school purpose or “is issued at the direction of a public school, a teacher, or any other employee” of a school, a school system or the State Department of Education. Before 2022 those were cumulative: a product had to be used primarily for school, issued at a school’s direction and designed and marketed primarily for schools. The 2022 rewrite made the first two alternatives and dropped the third, so a product that was not designed and marketed primarily for schools can now be an operator when a school agrees to use it for school purposes; it also added the FERPA condition, and it made a division of a larger company that serves education clients, and does not share covered information with its parent, an operator on its own.

Subsection (b) is the other half of the definition, and it is the reason this article has two halves. The Act “does not apply to a general audience” website, service or app, even if a student can use the same login on it. A product that families find and buy for themselves, with no contract with a school and no teacher directing students to it, is outside §4–131 entirely. It is not outside the law: federal children’s privacy rules, Maryland’s Online Data Privacy Act and, for a large company, the state’s Kids Code take over, as we explain below. But the line between the school product and the family product is drawn by the contract, not by the code, and the same codebase can sit on both sides of it.

Twenty-three kinds of covered information, nine kinds of identifier

Covered information is anything that, “alone or in combination with other information or material, is linked or could be linked to a student” in a way that would let someone at the student’s school identify the student “with reasonable certainty.” The statute then lists 23 kinds, and the list is the first thing we turn into a schema: educational records, first and last name, home address and geolocation, telephone number, email or other contact information, test results, grades and evaluations, special education information, criminal records, medical and health records, Social Security number, biometric information, socioeconomic information, food purchases, political and religious affiliations, text messages, student identifiers, search activity, photos, voice recordings, disciplinary information, “online behavior or usage of applications” linked or linkable to a student, persistent unique identifiers, and confidential information as defined by the state’s Department of Information Technology.

Persistent unique identifiers get their own list of nine: cookie identifiers, customer numbers, device identifiers, hashed email addresses, hashed phone numbers, identifiers generated through probabilistic methods, mobile ad identifiers, unique pseudonyms and user aliases. That list ends two arguments we have heard founders make. A hashed email is not anonymous under Maryland law, and neither is the random alias a product gives a student so that the teacher’s screen does not show a name. Both are covered information, and so is the analytics cookie on the login page.

Three duties, four prohibitions and ten ways to share

An operator must protect covered information from unauthorized access, destruction, use, modification or disclosure; implement and maintain reasonable security procedures and practices; and delete a school’s covered information “within a reasonable time” when the school or the school system asks. It may not knowingly engage in targeted advertising based on information acquired through the product, use the information to build a profile of a student except for a PreK–12 school purpose, sell a student’s information, or disclose covered information except in ten listed situations: in furtherance of the school purpose to a recipient bound by the same rules, for legal compliance, to take precautions against liability, in judicial process, for safety or the security of the service, to a service provider bound by contract to use the data only for the service, when federal or state law requires it, for legitimate research under the conditions the statute sets, to a state or local education agency for a school purpose, and, in a clause that on our reading swallows much of the rule, whenever the disclosure does not amount to targeted advertising, a non-school profile or a sale. That last exception is one reason districts lean on their own contracts, rather than on the statute, to say who may see their students’ data.

The service-provider route is where most of a startup’s architecture lives. Every hosting company, email sender, error tracker, analytics tool and AI model provider that touches covered information is a disclosure, and each needs a contract that prohibits it from using the data for anything except the service, prohibits further disclosure and binds it to the Act’s security and anti-advertising rules. A product with fourteen subprocessors needs fourteen such contracts, and a list of them that is current on the day a district asks.

The statute is also more permissive than its reputation in four places that matter to product design. An operator may use covered information for adaptive learning and customized student learning, and may recommend additional educational content within its own product, as long as no third party paid to influence the recommendation. It may use aggregated or de-identified information to develop and improve its products and to demonstrate their effectiveness, which is what an evidence study needs. It may use or disclose covered information beyond the listed exceptions, though never sell it, with clear and conspicuous notice to the student or a parent and the affirmative consent of a parent, or of the student if the student is at least 18. And nothing in the Act stops a student from downloading or exporting their own work. Two more clauses bind the future: a company that buys an operator inherits the Act’s obligations for the data already collected, and the Act may not be read to stop an operator from marketing educational products directly to parents, but only “if the marketing was not a result of the use of covered information” obtained through the school service. We come back to that sentence in the section on the family store, because it is the rule that decides how a company with two kinds of customer has to build its marketing list.

The list every district sends the state

The 2022 rewrite added a reporting duty that most vendors have never heard of. Under §4–131(p), as enacted by Chapters 163 and 164 of 2022, each county board must send the State Department of Education, by July 1 every year, three lists for the previous school year: the digital tools it approved, the digital tools “known to be used by educators,” and the digital tools “not authorized by the county board.” The Department must publish them as an online database every September 1. We looked for that database and could not find it. The Department’s website has been only partly online since a migration, our searches of its archived pages turned up the accessibility reports described in the next section but no privacy database, and two separate publishers of the Maryland code print §4–131 without subsection (p) at all, which looks like a codification slip rather than a repeal; no later act touched the section. The session law controls, so the duty stands, whether or not anyone can currently read the result. For a vendor, the practical point is that a teacher’s enthusiasm is not the end of the story: a product teachers use without approval is, by statute, a product the district is supposed to report as unauthorized.

What the state’s privacy council wants next

The same 2022 law re-created a Student Data Privacy Council, which runs until September 30, 2028, and required it to report by December 1, 2025 on best practices for protecting students’ data. The Council’s public minutes show where it was heading. Members surveyed the laws of Colorado, Illinois, Utah, New York and Montana, a West Virginia rule and the national data privacy agreement, and most wanted Maryland to codify explicit contract requirements for operators: the authorized uses of covered information, deletion at the end of the contract, the affiliated vendors that work on an operator’s behalf, and who pays, and what remedies follow, after a breach. At its last meeting before the report, on November 10, 2025, the Council added a recommendation that operators carry cyber insurance for data breaches, agreed that local boards should adopt or update data governance policies, and heard a Department staff member caution that adopting the Student Data Privacy Consortium’s national agreement as Maryland’s template “might go against the State’s fair contracting practices because of vendors’ role in creating the Consortium.” The Council approved its report by a roll-call vote of ten to none on November 10, 2025, and a 2026 fiscal note says it has been issued, but we could not find a published copy, and no 2026 law came out of it. A founder who builds deletion at the end of a contract, a breach notice to families and proof of cyber insurance into the product now is building what Maryland’s own advisers have asked the legislature to require.

The laws that apply when there is no school contract

The family side of the business answers to different statutes. Maryland’s Online Data Privacy Act, Commercial Law §14–4701 and the sections after it, took effect on October 1, 2025 and has applied to processing since April 1, 2026. It reaches a company that controls or processes the personal data of at least 35,000 Maryland consumers, not counting data used only to complete a payment, or of at least 10,000 if more than 20% of its revenue comes from selling personal data. It exempts government bodies and data “regulated by” FERPA, so a district’s records held under a school contract stay with §4–131, but it does not exempt the company or its family customers. Where it applies, it is one of the strictest state privacy laws for young users: a controller that “knew or should have known” a consumer is under 18 may not sell that consumer’s data or use it for targeted advertising, a child’s data is sensitive data that may be collected only when “strictly necessary,” and complying with COPPA’s parental consent rules counts as consent for a child under 13. Chapter 874 of 2026 rewrote the definition of sensitive data from July 1, 2026. Our article on the Act and online stores walks through the thresholds.

The Maryland Kids Code, Commercial Law §14–4801 and the sections after it, has been in effect since October 1, 2024, and it reaches fewer edtech companies, but the ones it reaches it reaches hard. It applies to for-profit companies doing business in Maryland with more than $25 million in revenue, data on 50,000 or more consumers, households or devices, or at least half their revenue from selling data, whose product is “reasonably likely to be accessed by children,” which here means anyone under 18. It has no FERPA exemption. A covered company must complete a data protection impact assessment for each such product, due by April 1, 2026 for products already on offer that would still be offered after July 1, 2026, and before launch for new ones; set high-privacy defaults; keep profiling and precise geolocation off by default; and notify both the child and a parent before letting anyone other than the parent monitor the child’s online activity, which, on our reading, includes a teacher using a classroom-monitoring feature. Penalties run up to $2,500 per affected child for negligent violations and $7,500 for intentional ones. The trade association NetChoice challenged the law in February 2025; the federal court in Baltimore denied the state’s motion to dismiss on November 24, 2025, noting that “the Kids Code remains in effect,” let NetChoice add a Commerce Clause claim in July 2026, and is now in discovery, with no injunction in place.

Two older rules finish the picture. Maryland’s breach notification law, Commercial Law §14–3504, gives a business 45 days to notify affected residents after it discovers a breach, requires notice to the Attorney General first, and says that a vendor holding data it does not own must tell the owner “as soon as practicable” and no later than 10 days after discovering the breach, at no charge; for an edtech company, the owner is the district. And the state’s student records regulation, COMAR 13A.08.02, lets a school disclose records without consent only in listed cases, starting with school officials who have a legitimate educational interest, and requires a record of most other disclosures, including any onward disclosure made on the school’s behalf, which is why a district will ask a vendor for logs it can attach to its own.

The accessibility test waits for your login, and AI skips it until 2028

Most writing about edtech accessibility starts with the federal rule, and we will get to it. In Maryland it is the second rule you meet, not the first. Education Article §7–910 governs every digital tool a county board buys, and since Chapter 215 of 2022, sponsored by Senator Craig Zucker, it has read like a procurement checklist written by someone who had been burned. A “digital tool” is an online platform, an online course, software and operating systems “directly connected to student instruction,” digital content or other digital technologies, and every grant and procurement contract for one must require “equivalent access” for students with disabilities, including blindness, measured against the federal Section 508 standards. The statute says what that means: a student who cannot see must be able to “receive, use, and manipulate information and operate controls” through keyboard controls, synthesized speech, braille or other means, “with substantially equivalent ease of use.”

The obligations then fall on the vendor in a sequence a product team can plan around. Since September 1, 2023, every invitation for bids or request for proposals for a digital tool issued by a local school system must require the vendor to submit an accessibility conformance report that includes a Voluntary Product Accessibility Template, explaining how the software, its electronic content and its support documentation conform to the most recent Section 508 standards. Each school system must evaluate a tool before buying it, through an employee or contractor who specializes in accessibility and the Web Content Accessibility Guidelines or is a blindness specialist. Since October 1, 2024, a school system must select, among products with pedagogical value, the one that “best meets the equivalent access standards,” and since 2022 it has been barred from approving a contract for a tool that fails them unless it has notified the state, which consults the Department of Information Technology and the Department of Disabilities and may let it buy the best of a weak field. The Department asks districts to allow up to two months for that.

What happens after the sale is the part that should change how a startup writes its contract. The procurement contract must require the vendor to indemnify the school system for liability and costs if the tool fails these requirements. If, within 18 months of buying a tool, the school system finds that it fails, it sends the vendor a written notice, and the vendor must fix the tool “at the vendor’s expense” within a time the two agree. A vendor that does not is subject to a civil penalty of up to $5,000 for a first offense and up to $10,000 for a later one, and must indemnify the board. Separately, anything a county board develops or buys and makes available to its students, or to the public online, must comply with “the most recent version” of the Web Content Accessibility Guidelines, which today is WCAG 2.2. The state’s regulation, COMAR 13A.06.05, still incorporates WCAG 2.1, as adopted in 2019, and the Department wrote in December 2025 that “no action is expected” on 2.2 because the federal rule adopted 2.1; a vendor that tests against 2.2 satisfies both readings.

What the state’s own reports show

The statute also makes the results public. Every October 1, each school system reports to the state on the accessibility of the digital tools it bought in the previous fiscal year, and the Department publishes the files and summarizes them for the Governor and the General Assembly every December. The Department’s website has been partly offline since its migration, but the district files survive in the Internet Archive, and we read the two that matter here.

Baltimore City Public Schools’ report for fiscal 2024 lists 114 named digital tools, from Achieve3000 and Canvas to Zearn and Zoom. For every one of them the district certified that its bids carry the accessibility notice and that its contracts require the vendor’s indemnity, and for every one it answered “No” to whether a VPAT was on file and “No” to whether the tool conforms to the state’s standard. Fifty-five tools had an accessibility evaluation on file. The other 59 carry the same comment: “Login credentials not provided by the vendor. Evaluation is pending evaluation credentials.” In other words, more than half of the City’s list was waiting on vendors to hand over a test account. Baltimore County Public Schools reported only its eight new purchases for the year: seven had an evaluation on file, five conformed, three did not, and none of the eight contracts carried the indemnity clause.

The district rows come from the two school systems’ fiscal 2024 files under §7–910(i), which we retrieved from the Internet Archive; the statewide rows come from the Department’s December 2024 and December 2025 reports under §7–910(j) and count every district’s tools, duplicates included; the 2025 report gives the fiscal 2025 evaluation rate as 85% in one table and 87% in another. The fiscal 2025 files for Baltimore were not online when we checked, so the City’s numbers may have moved since.

Statewide, the trend is steep and in the right direction. In fiscal 2025 the 24 school systems reported 2,201 tools; 69% conformed, up from 49% a year earlier and 23.2% the year before that, the share with a VPAT on file rose from 69% to 88%, and the share of contracts with the indemnity clause rose from 29% to 64%. Against the statewide VPAT rate of 69% for fiscal 2024, the City’s zero for the same year stands out. The same report records 317 notices from districts that intended to buy a tool they knew did not meet the standard: 224 citing undue burden, 30 because no accessible product existed, 13 because an accessible one would fundamentally alter the activity, and 50 for more than one reason. For a vendor, the 317 is the market speaking. Three hundred and seventeen times in one year, a Maryland district wanted a product badly enough to file paperwork explaining why its accessibility was not good enough, and in each of those cases an accessible product with the same teaching value would have been the one the statute told the district to buy.

The three-year pause for AI

There is one large exception, and it is new. The Excellence in Maryland Public Schools Act, Chapter 237 of 2025, added a subsection (a–1) to §7–910: beginning with the 2025–2026 school year and through the 2027–2028 school year, the section “does not apply to the procurement and use of a digital tool that utilizes artificial intelligence” to support student learning, using the state’s procurement definition of AI. The change took effect on July 1, 2025 and lapses at the end of June 30, 2028, after which the original text returns. The fiscal note gives no reason for it. The Department told districts in August 2025 that each school system decides for itself whether a tool “utilizes” AI, that such a tool needs neither an accessibility review nor a notice to the state until the pause ends, and that the fiscal 2026 reports, due on October 1, 2026, would add a list of the AI tools each district bought.

We would not build a product around that pause. It covers only the state statute, not the federal ADA rule that applies to the same districts from April 2027, it ends with the 2027–2028 school year, and a district that has spent three years asking every other vendor for a conformance report is unlikely to forget to ask the AI vendor. For a product team, all of this resolves into a handful of records. The conformance report has to exist before the first bid, name the standard and version it was tested against, and be regenerated when the interface changes. Keyboard navigation, screen reader labels, captions, color contrast and braille-display compatibility are acceptance criteria with a test that runs on every release. A district evaluator needs a working test account with realistic data on the day the product is submitted, not after a support ticket. And the date each district bought the product matters, because the 18-month window for a notice runs from it.

Since June, a state rubric decides whether an AI tool gets a pilot

If a product uses a language model, Maryland now has a second gate for it, and it is newer than everything above. The Artificial Intelligence Ready Schools Act, Chapter 634 of 2026, was signed on May 26 and took effect on June 1, 2026, adding a new subtitle to the Education Article. It requires the State Department of Education to publish guidance on AI for students, educators, administrators and parents and to review it every year; requires every local school system to adopt an AI policy aligned with that guidance within 120 days of its release and to name a central-office coordinator for AI; requires the Department to publish a rubric for evaluating AI tools, which each local school system “shall use” when reviewing and selecting them; and requires school systems to buy AI tools consistent with the state’s procurement rules for artificial intelligence in Title 3.5, Subtitle 8 of the State Finance and Procurement Article. Morgan State University or another four-year university may help districts score tools on request. The Maryland Association of Counties told its members in September that most school systems were likely due to have their policies in place that month.

What the rubric asks

The Department had already written the rubric. Its AI framework, “Maryland’s Framework for Responsible AI in Education,” went to the State Board of Education on February 24, 2026 with an AI Tools Evaluation Rubric among its attachments, and the rubric is specific enough to design against. It scores a tool from one to four on nine criteria: data privacy and security, instructional and cognitive presence, bias, equity and access, human in the loop, implementation and piloting, professional development, interoperability with the district’s learning management system, technical performance and environmental impact. The top score for privacy describes a dedicated education tier whose data “never trains global models”; the top score for human oversight describes teacher control over AI feedback and grades; the top score for interoperability describes an LTI launch from Canvas or Schoology with roster sync. Before any pilot, four questions must all be answered yes: the vendor contractually agrees not to use district data, student or staff, to train its foundational models; the tool clearly labels AI-generated content; users have a visible way to flag “hallucinations,” inaccuracies or inappropriate content; and the district can request the permanent deletion of user data. A tool that scores 24 to 28 points is approved for a pilot, 18 to 23 is conditionally approved with contract addenda, and below 18 is denied. The bands stop at 28 although nine criteria allow 36, and the lowest privacy level cites Illinois’s student privacy law, both signs of a first edition that districts will interpret as they go.

The framework around the rubric says more about vendors than most edtech contracts do. Contracts should limit data use, retention and deletion and “prohibit advertising, profiling, or unrelated model training”; vendors must document how their AI works, what data it collects and keeps, whether any data is used for training, its known limitations and its accessibility and anti-bias measures; districts need audit rights, breach terms and deletion at termination; a person must review any AI output that affects placement, grading, discipline or access to supports; and “collection of biometric, behavioral, or location data is prohibited unless legally required and educationally justified.” Combined with the state’s three-school-year pause on accessibility review for AI tools, the result is a strange incentive: an AI feature escapes Maryland’s accessibility statute until mid-2028 but walks straight into a privacy and oversight rubric that most conventional software never faces.

What Baltimore’s districts say

City Schools updated its academic AI guidance on May 29, 2025, building it on the national TeachAI toolkit. It currently suggests Microsoft Copilot, Google Gemini and NotebookLM, used while signed in with a district account; tells staff and students not to enter personal information into AI tools “without City Schools’ approved data privacy agreements”; commits that AI “will not be used to monitor classrooms for accountability purposes”; and labels assignments on a five-level scale of acceptable AI use, under which students at every level that allows AI must include links to their AI chats with their work. We found no published AI policy from Baltimore County, though a summary of its September 8, 2026 board meeting says staff will draft one aligned with the state guidance. The State Board presentation also mentions a pilot of Khan Academy’s Khanmigo with Dorchester and Baltimore County schools.

Independent ratings point the same way. Common Sense Media rated Gemini for K–12 users “high risk” on January 9, 2026, finding that it “lacks age-appropriate responses and adequate mental health support,” and rated AI teacher assistants, including Gemini in Classroom, Khanmigo, Curipod and MagicSchool, moderate risk after finding bias in the behavior plans they wrote. Digital Promise launched a Responsibly Designed AI certification on April 7, 2025 that asks for evidence of transparency about data, a breach response plan, bias monitoring, labeling of generative features and the ability of educators to override AI output, which is very nearly Maryland’s pre-pilot checklist.

The federal and out-of-state rules an AI tutor runs into

Federal policy encourages AI in schools and constrains how it is built. Executive Order 14277 of April 23, 2025 set up a White House task force and a Presidential AI Challenge; the Education Department told districts on July 22, 2025 that federal grant money may pay for AI tools and AI-enhanced tutoring, made AI a priority for its discretionary grants in a final rule effective May 13, 2026, and on August 20, 2026 told districts to prefer products backed by randomized controlled trials and vendors to “publish rigorous, independent evaluations.” The amended COPPA Rule, as we explain below, treats disclosing children’s data to train AI as not integral to a service and therefore as needing separate parental consent. In September 2025 the Federal Trade Commission ordered seven companies, including OpenAI, Meta, Alphabet and Character Technologies, to report how their AI companions affect children. And the December 11, 2025 executive order on a national AI policy framework, which aims to preempt “onerous” state AI laws, expressly does not propose preempting state child safety protections.

That matters because states are writing rules for chatbots that talk to children, and an AI tutor can fall inside them. California’s SB 243, in force since January 1, 2026, covers chatbots that can meet a user’s social needs across sessions, does not exclude education, and requires a notice that the bot is not human, a published self-harm protocol and, for known minors, a break reminder every three hours. California’s SB 1119, signed on September 10, 2026, goes further from July 1, 2027: verify age or treat every user as a child, keep memory and push notifications off and sessions to an hour by default, and stop a bot from claiming feelings or flattering excessively. New York’s companion law requires a “not human” notice at the start and every three hours for bots that remember earlier sessions and ask unprompted emotional questions. Maryland’s own companion-chatbot bill, HB 952 of 2026, passed the House and died in the Senate. The practical design rule we take from all of it is simple: a tutor that helps with fractions, labels its answers as AI, keeps no relationship memory and asks no unprompted questions about feelings is outside most of these laws, and a tutor with a persona and a memory can fall inside all of them.

The federal layer: what changed for edtech vendors in 2026

Federal law sits underneath everything above, and in the last eighteen months it has moved from guidance toward enforcement. Four changes matter most to a company that sells to Baltimore schools: the amended children’s privacy rule is now in force, the Federal Trade Commission has finalized an order against an edtech company over a breach, the Education Department has started writing to vendors directly, and the federal accessibility deadline for school districts is now set for April 2027.

FERPA: the vendor stands in the school’s shoes

A district that shares education records with a vendor without parental consent normally does so by treating the vendor as a “school official,” and the federal regulation, 34 CFR 99.31(a)(1)(i)(B), allows that only if the vendor performs a service the district would otherwise use employees for, is “under the direct control” of the district “with respect to the use and maintenance of education records,” and is bound by the rule against passing records on. A vendor that rediscloses records improperly can be barred from receiving personal information from that district “for at least five years.” In product terms, direct control means the district, not the vendor, decides who sees what, for how long and when it is deleted, and the product has to give the district the controls to do that.

The Department’s Student Privacy Policy Office has also started to address vendors directly. On May 12, 2026 it wrote to the chief executive of Instructure, the maker of Canvas, after cybersecurity incidents, saying that “violations of FERPA likely occurred,” criticizing the company’s agreement to pay a ransom, and demanding within two weeks the dates, the students affected, the kinds of data, the remediation and the company’s communications with districts, along with a long list of documents: its information security, data classification, access control and vendor risk policies, its incident response plan and tabletop exercise logs, access reviews, patch and change records, network diagrams, its SOC 2 Type II report, ISO 27001 certificate and third-party penetration tests, and its “AI Usage Policies.” A three-person company will not get that letter, but its larger district customers now know what to ask for, and that list is the most specific description of a district-ready security file we have seen from the federal government.

The Department’s other FERPA actions in 2026 were aimed at states and districts, among them a finding against California in January and an enforcement action, announced with the Justice Department in July, against Anne Arundel County Public Schools, next door to Baltimore. Both turned on information about students kept from their parents, and the California finding singled out requests to a student information system vendor for features that would hide changes to a student’s name and pronouns from the parent portal. Whatever one thinks of the underlying dispute, the product lesson is narrow and neutral: a parent has a federal right to inspect a student’s education records, so every student-linked record a product keeps for a district should be exportable on the district’s request, and a product should not create categories of record designed to sit outside that right.

COPPA: the amended rule has applied since April 22

The Federal Trade Commission published its amendments to the Children’s Online Privacy Protection Rule on April 22, 2025; they took effect on June 23, 2025, and companies had until April 22, 2026 to comply. The changes that matter most to an edtech product are concrete. Parents must be able to consent to collection and use without consenting to disclosure to third parties, and where disclosure is not integral to the service, the company needs a separate consent for it; the Commission’s explanation names disclosures “for advertising purposes, or to train or otherwise develop artificial intelligence technologies” as not integral. The company needs a written information security program with a named coordinator, an annual risk assessment and regular testing, and a written data retention policy, published in its privacy notice, that says what is kept, why and when it is deleted, because children’s data “may not be retained indefinitely.” The notice must now name the identities and specific categories of the third parties that receive children’s data. And the rule adds new ways to obtain a parent’s consent, including knowledge-based questions and a government ID matched to a face and then deleted, and it loosens the payment-card method so that the transaction no longer has to be a charge.

The Commission declined to codify the exception that lets a school consent on parents’ behalf, saying it would continue to enforce COPPA in edtech “consistent with its existing guidance” rather than risk conflicting with future FERPA rules. That guidance says a school may consent only where the data is collected “for the use and benefit of the school, and for no other commercial purpose,” that the company must give the school full notice of its practices, let it review and delete the data and stop collection, and that a company should not write into its terms that the school is responsible for complying with COPPA. A product with a school side and a family side therefore needs two consent paths: school authorization for the classroom, limited to the school’s purposes, and verifiable parental consent for anything a family signs up for, with any third-party disclosure behind its own switch.

The FTC’s order against Illuminate

On June 5, 2026 the Commission voted 2–0 to finalize an order against Illuminate Education, a company whose software schools use for assessment and student data. The complaint described a breach of 10.1 million students’ data through the credentials of a former employee, used three and a half years after the employee left; vulnerabilities a vendor had flagged in January 2020; student data kept in plain text until at least January 2022; and notices to some districts that took nearly two years. The order requires the company to minimize the data it collects, delete what it no longer needs, publish a data retention schedule, run a comprehensive security program and tell the Commission whenever it reports a breach to other governments, with civil penalties of up to $51,744 for each later violation. Every one of those allegations maps onto an ordinary engineering habit: revoking access on the day someone leaves, encrypting data at rest, writing down how long data is kept, and knowing within hours which districts a breach touched. In 2023 the Commission’s order against Edmodo had already banned a company from requiring more data than a classroom activity needs and criticized the practice of “outsourc[ing] compliance responsibilities to schools.”

The ADA rule now names April 2027

In April 2024 the Justice Department adopted a rule under Title II of the Americans with Disabilities Act requiring state and local governments, including public school districts, to make their web content and mobile apps conform to WCAG 2.1 Level AA, including content and apps they provide “directly or through contractual, licensing, or other arrangements,” which is a description of third-party edtech. The rule deliberately has no exception for course content behind a password. On April 20, 2026 an interim final rule moved the compliance dates by a year: to April 26, 2027 for public entities serving a population of 50,000 or more, which covers both Baltimore school systems, and to April 26, 2028 for smaller ones. The Department said it may reopen the rule’s substance. In the meantime, the Education Department’s Office for Civil Rights showed what enforcement looks like in a July 2026 agreement with a Pennsylvania school district: adopt WCAG 2.1 AA within 30 days, inventory every web address “including those on all vendor-hosted or third-party-hosted sites,” fix the barriers by March 31, 2027 or seek to replace the vendor, require bidders to commit to remediation for the life of the contract, and verify “any vendor-offered accessibility assessment” independently. The current template for that assessment is VPAT 2.5Rev, published on April 24, 2025 in four editions; a completed VPAT is what the industry calls an accessibility conformance report.

Evidence: the tiers, and the Baltimore institution that rates them

The Every Student Succeeds Act sorts evidence into four tiers: strong, from at least one experimental study; moderate, from a quasi-experimental one; promising, from a correlational study with statistical controls; and “demonstrates a rationale,” which needs a research-based logic model and “ongoing efforts to examine the effects.” The fourth tier is the entry ticket for a new product, and the Education Department’s guidance says what it needs: a logic model “informed by research or evaluation findings.” School improvement money under section 1003 of Title I can only buy programs in the first three tiers, and the Department’s guidance expects a sample of at least 350 students across more than one site for the top two.

One of the best-known independent rating sites is in Baltimore. Evidence for ESSA is run by the Center for Research and Reform in Education at Johns Hopkins University, and its standards, revised in June 2026, require at least two schools and 350 students for a strong or moderate rating and at least two teachers and 30 students in each condition for a promising one; it does not rate the fourth tier, does not accept measures the developer made, requires at least twelve weeks and reviews only American studies. A developer submits a study by email, and the site says plainly that “there is no such thing as official ‘ESSA Approval.’” When we pulled its public data on October 17, 2026 it listed 1,292 programs: 81 rated strong, 38 moderate, 77 promising, and 1,094 with no qualifying evidence. For a startup, that ratio is the argument for building the data collection an evidence study will need into the first version of the product, with the consents and the de-identification the study will require, rather than trying to reconstruct it in year three.

Federal money comes with its own purchasing rules

When a district pays with federal grant money, the Uniform Guidance at 2 CFR part 200 applies on top of the district’s own rules. Its thresholds follow the federal acquisition regulation, which raised the micro-purchase threshold from $10,000 to $15,000 and the simplified acquisition threshold from $250,000 to $350,000 on October 1, 2025, although a district may set lower limits and City Schools’ own rules are lower. Two provisions matter more to a small vendor than the numbers. A contractor that helps “develop or draft specifications” for a procurement must be excluded from competing in it, so a founder who writes a district’s requirements for it as a favor may have written itself out of the bid. And above the simplified threshold the district must analyze the cost or price of every purchase, which is easier to pass with a published price than with a quote that changes by customer.

What the vendors on Baltimore’s lists show a district, a parent and a machine

Neither Baltimore school system publishes its full list of approved tools, and the state database the 2022 law requires was nowhere we could find. The most complete public record of what Baltimore’s schools actually buy is the accessibility reporting that §7–910 requires, so we used that. We combined the tools Baltimore City and Baltimore County reported to the state for fiscal 2023 and 2024 with the 40 entries on Baltimore County’s 2026–2027 Innovation HUB list, removed hardware, bundles and duplicates, and collapsed the 278 names into 169 vendor families, 75 on City lists only, 69 on County lists only and 25 on both. We drew 50 at random and read each company’s homepage, pricing, privacy, accessibility, security, family and integration pages, its robots.txt and its llms.txt, as of October 17, 2026.

Seven of the 50 could not be read as working products. Two domains no longer resolve, one tutoring organization’s parent site shows an expired website notice, one tutoring company’s site returned server errors, one product’s domain now carries casino review posts in German, one site blocked every automated reader we tried, and one assessment platform was retired on September 1, 2026 with a notice that “student data and assessment history can no longer be accessed.” Two more had been absorbed by another company or were being wound down. These are tools that Baltimore’s school systems told the state they bought in fiscal 2023 or 2024, which is the strongest local argument we know for writing deletion, export and transition terms into every contract.

The 43 we could read describe an industry that has learned to say “FERPA” but not “Maryland.” Twenty-four name FERPA and 23 name COPPA. Not one names Maryland’s Student Data Privacy Act or §4–131, and only one mentions Maryland in a student-privacy context at all; the one Maryland school law any vendor cited was the accessibility statute, by its 2022 bill number. Five name the Student Data Privacy Consortium or its national agreement. Ten claim a SOC 2 report and six an ISO 27001 certificate, and fourteen have a security or trust page.

Accessibility, which Maryland requires in every bid, is where the gap is widest. Thirteen of the 43 have an accessibility statement, seven link to or mention a VPAT or conformance report, and seven name a WCAG version: 2.1 on four sites, 2.2 on two and 2.0 through 2.2 on one. That squares with the City’s fiscal 2024 report, in which none of the 114 tools had a VPAT on file: the reports exist for some products, but most vendors do not publish them where a district reviewer, or a parent, could find them, and nearly half of the companies we could read are large publishers and platforms with the staff to do it.

Eighteen advertise AI features, and only two of those 18 say anything about whether student data is used to train AI models: one says plainly that it does not use personal data for training, and the other says it does not train on raw voice recordings but does use de-identified data. Fourteen claim evidence under ESSA, at tiers from II to IV, though one site claims Tier III on its homepage and Tier II on its pricing page. Eleven publish a price, two of them per student; sixteen offer a free teacher tier; and seven let a family buy directly, from a $12.95 monthly family plan for a math program to a publisher’s consumer store.

The machines get mixed signals. Twelve publish a real llms.txt, two block one or more AI crawlers in robots.txt, and 25 carry structured data on the homepage, but only three describe their product as a SoftwareApplication, the type Google uses for software. A few vendors did the dull things well. Seesaw’s privacy page names the consortium, FERPA, COPPA and SOC 2, links a trust portal and backs an accessibility statement that cites WCAG 2.2 with a VPAT. WeVideo publishes prices, links a VPAT and says in one sentence that it does not use personal data to train AI models. Happy Numbers, a small math company, publishes per-student prices, a separate student privacy policy and a family plan. And Lightspeed Systems is the only vendor in the sample that cites a Maryland school law by name, the 2022 accessibility bill, on its page about state compliance laws.

What 72 independent edtech companies show a district, a parent and a machine

To see whether Baltimore’s picture is local or national, we drew a second sample from the one list where every product is, by construction, something schools actually connect: Clever’s public App Gallery, which on October 17, 2026 listed 1,810 apps on 1,551 distinct websites. We removed public bodies, single-district instances, apps hosted on app stores and the large publishers and platforms, which left 1,392 independent vendor domains, and drew 80 of them at random. Then we read each company’s homepage, pricing, privacy, accessibility, security and family pages, its robots.txt and its llms.txt, the same way a district reviewer, a parent and a crawler would.

The first finding is about time. Eight of the 80 listings no longer lead to a working product: three domains no longer resolve, two return “not found,” one returns a server error, one product was retired into the charter network that built it, and one domain now belongs to an unrelated construction-software company. Nine more have moved, rebranded or pivoted, one of them into designing vehicle wraps. That is 17 of 80, or 21%, with some kind of churn in a list that schools use to connect students to software today, and it is the plainest case we know for building a district’s deletion and transition rights into the product rather than trusting the company to be there in three years.

The other 72 are the base for everything below, and most of them sell the way enterprise software sells. Only 19 publish any price at all (26%), and only 10 (14%) publish a price for a school or classroom license; seven price per student, eleven have a free teacher tier and six are entirely free, while 27 (38%) make “request a demo” the main button. Fifteen (21%) let a parent or an individual sign up or buy directly, which is the family store this article argues for. The best published prices were refreshingly specific: one parent-communication app charges $3, $5 or $7 per student a year, one special-education data tool charges $72 a year for up to 18 students and $4 per student after that, and one grading tool charges $2.50 per student a year plus $999 per building.

The privacy pages were thinner than we expected from products that schools have already connected. Thirty-four sites (47%) name FERPA and 36 (50%) name COPPA, but 28 (39%) name neither on the pages we could read. Seven name a specific state student privacy law, most often California’s and New York’s, and exactly one names Maryland in a privacy context. Three mention the Student Data Privacy Consortium or its national agreement, one publishes a state-by-state list of signed agreements (sixteen states, not including Maryland), and seven offer a data privacy agreement on request or publish their own template. Four still display the Student Privacy Pledge, which its sponsor, the Future of Privacy Forum, retired on April 25, 2025, and two still use the old name of the standards body that now calls itself 1EdTech.

Security and accessibility were thinner still. Five companies (7%) claim a SOC 2 report of their own, and four more present their cloud provider’s certificate in a way that reads like their own. Eleven (15%) have an accessibility statement and five (7%) publish a VPAT or accessibility conformance report, three of them against WCAG 2.1; in Maryland, where §7–910 has required that report in every school system’s request for proposals since 2023, the other 67 would have to produce one before their first bid. Three (4%) make a claim about evidence under ESSA, and only one names a tier. Twenty-two (31%) advertise AI features, and only four of those 22 say that student data is not used to train AI models.

The machine-readable picture was mixed. Seventeen percent name an AI crawler in their robots.txt, but only three (4%) block one, which suggests that most edtech companies would rather be found by AI assistants than hidden from them. Fourteen (19%) publish a valid llms.txt, while ten more return a normal web page with a success status at that address, which a careless check would count as having one. Thirty-four (47%) carry any structured data on the homepage, almost all of it a basic Organization or WebSite record, and five (7%) describe the product itself as a SoftwareApplication. A few companies stood out for doing the dull things well: Bloomz publishes per-student prices, a VPAT against WCAG 2.1 and an llms.txt; ReadWorks pairs free teacher accounts with a SOC 2 report, a public conformance report and an ESSA Tier 3 certificate from Digital Promise; AEGIX runs a trust center with a dated VPAT and a public list of its state agreements; DataClassroom publishes its prices and offers to sign a district’s agreement; and ClassCloud publishes a data processing addendum that promises no AI training on student data.

One caveat runs through all of this. These are products that already connect to Clever, so they are probably more ready for districts than edtech in general, and we read about fifteen pages per site from a single location on a single day; a policy that lives behind a login or in a PDF we did not reach is counted as missing, which means some of these numbers, especially for LTI, SOC 2 and VPATs, are undercounts.

The other store: selling to families

Everything so far describes the district door, which is slow, paperwork-heavy and usually worth it. The family door is faster, and for many Baltimore edtech companies, tutoring businesses and curriculum authors it is where the first revenue comes from. It runs under different law, and the most useful thing a founder can do is build the two doors as two doors from the first day.

Two doors, one rule

Maryland’s student data law does not apply to a general audience product that families find and buy on their own, and it explicitly allows an operator to market educational products to parents, but only if “the marketing was not a result of the use of covered information obtained” through the school service. Read as a data model, that sentence requires two customer tables, two marketing lists and two sets of identifiers that never meet. A parent whose child uses the school version can buy the family version, but the company may not find that parent through the school version, recommend the family plan inside it, or use the child’s classroom data to decide what to sell. The family side then answers to COPPA, with a verifiable parental consent before a child under 13 creates an account and a separate consent for any disclosure that is not integral to the service, and, if the company is large enough, to Maryland’s Online Data Privacy Act, with no sale of data and no targeted advertising for any user the company knew or should have known was under 18.

Checkout can do double duty. Under the amended COPPA Rule, a payment card or other online payment system used “in connection with a transaction,” with notice of each transaction to the account holder, counts as verifiable parental consent, and the 2025 amendments removed the word “monetary,” so the transaction no longer has to be a charge. A family store whose first step is a parent paying, or authorizing, with their own card has collected its consent in the same motion, provided it gives the notice the rule requires and keeps the record.

What a $10 workbook earns in each store

Teachers Pay Teachers, the marketplace where teachers sell classroom materials, changed its terms for sellers on October 1, 2026. Its Premium membership closed to new sellers and is being replaced by an Exclusive membership at $59.95 a year that pays 80% of each sale until January 1, 2027 and 77.5% after that, with no transaction fee on orders of $3 or more and 15 cents per resource below that. The price of that rate is exclusivity: an Exclusive seller agrees that “any resources that are eligible to be sold on TPT must be sold exclusively through TPT,” which covers digital educational resources, memberships that include them and licenses for them, and the company’s own website may promote them but must send buyers to the marketplace to pay. The Basic membership, which allows selling elsewhere, costs $29 once and pays 55% of each sale minus 30 cents per resource. Our own store’s figures use Stripe’s published rate for domestic cards, 2.9% plus 30 cents.

The arithmetic favors the company’s own store on every sale, and the marketplace on every sale the company would not have made on its own. A marketplace brings buyers, handles sales tax as a marketplace facilitator and asks nothing of the seller’s engineering team; a company’s own store has to bring its own traffic, register to collect Maryland sales tax and file returns, and pay for the store. The choice is a business decision rather than a technical one, but the Exclusive membership has made it binary for digital resources: a company that wants its own family store cannot also be an Exclusive seller of the same resources.

One subscription, four tax answers

Maryland taxes the same product differently depending on who buys it, which is another reason the family store and the school product should not share a checkout. A district’s purchase is normally exempt, because sales to the state and its political subdivisions are exempt, and the district will tell you which exemption it relies on. A Maryland nonprofit charitable or educational organization, such as a private school with the Comptroller’s exemption certificate, buys tax free, and the seller must keep the certificate on file for at least four years. A business that buys software as a service for its own enterprise systems, such as a tutoring company buying a scheduling tool, pays 3% under the tax on data and information technology services that began on July 1, 2025. A parent who buys the same subscription for personal use pays 6%, because software bought for individual use is a digital product, and where two rates could apply, the higher one wins; a downloadable workbook is a digital product at 6% too. Our article on the 3% tax covers the enterprise side and our article on tutoring explains why live sessions are generally not taxed.

App stores have rules for children, too

A family product that is an app answers to two more rulebooks. Apple’s guidelines, last updated on October 16, 2026, forbid apps in its Kids Category from linking out of the app or offering purchases except behind a parental gate, restrict third-party analytics and advertising, forbid advertising based on data from Apple’s ClassKit or from schools, and require consumer and family purchases of digital content to use Apple’s in-app purchase, although apps on the United States storefront may now also link to other ways to pay. Google Play’s Families policy requires a declared target audience, allows only self-certified advertising libraries for children and bars apps aimed only at children from sending the device’s advertising identifier; its Teacher Approved program, which earns a badge and eligibility for the Kids tab, was still running in October 2026.

How families, and their AI assistants, find you

Search has narrowed the special treatment it used to give education sites. Google retired the Course info and Learning video rich results on September 9, 2025, removed its practice problems documentation on January 6, 2026, and stopped showing FAQ rich results on May 7, 2026; its flashcard format, Education Q&A, is still live. For software, Google’s SoftwareApplication markup needs a name, a price in its offers, even if the price is zero, and either a rating or a review. Google says AI Overviews and AI Mode need no special markup and that Google Search ignores llms.txt, which will “neither harm nor help.” The assistants are separate crawlers: OpenAI’s OAI-SearchBot fetches pages for ChatGPT search while GPTBot collects training data, Anthropic separates Claude-SearchBot from ClaudeBot, and Google’s Google-Extended token controls Gemini’s use of a site without affecting Google Search, so a company can block training and still be found.

What families and founders type is a better guide than any markup. Google’s autocomplete on October 17, 2026 suggested “coppa april 22 2026,” “does coppa apply to schools,” “what is a vpat for software,” “accessibility conformance report vs vpat,” “ada title ii deadline extension,” “tpt school access going away,” “does maryland tax saas” and “baltimore city public schools vendor registration requirements,” and almost every top result we read for those questions was a law firm, a vendor or a government page, several of them out of date. A Baltimore edtech company that answers its own customers’ versions of those questions on its own pages, in plain text with a date on it, is giving both the search engine and the assistant something to quote.

A family store that sells the next step

One of the clearest family stores we have looked at recently is not American. edtalk.lt, a Lithuanian online tutoring company, runs a store whose public catalog listed 65 products on October 17, 2026, 59 in mathematics and six in English, priced from €2.38 to €29.99 with a median of €16.99. They come in three kinds that a Baltimore company could copy line for line: 21 downloadable PDFs, such as a mock entrance exam for Vilnius Lyceum, one of the country’s most selective schools, at €16.99 and a set of ninth-grade mock exams at €29.99; 20 printed workbooks by well-known Lithuanian authors, sent after the sale, such as a fourth-grade test book at €2.38; and 24 bookable lessons, such as a one-lesson diagnostic of a student’s level at €25. Thirty-one of the 41 downloads and workbooks are paired with a lesson in which a tutor goes through the child’s mistakes with them, so the store sells the next step as well as the material. In Baltimore the same catalog would be a set of practice tests built around the state’s MCAP assessments and the admission tests the area’s independent schools use, a printed workbook line and a booking page, with the session as the upsell, which is exactly what our Online Store package is built to do.

A note for our readers in Lithuania

Part of our team works from Lithuania, and the comparison is instructive because the two systems put the same decisions in different places. In Maryland the district is the gatekeeper, with a procurement office, a privacy agreement and now a state rubric. In Lithuania the school is. The Ministry of Education, Science and Sport published guidelines on the use of artificial intelligence on January 16, 2026 that advise each school to keep its own list of permitted AI tools, approved by the principal’s order, and bar it from listing tools that use emotion recognition, biometric tracking or social scoring, or that say nothing about data protection. The ministry’s own survey found that about 68% of teachers use AI, nearly all of them ChatGPT, and the National Agency for Education is renewing some 330,000 student Microsoft 365 licenses with Copilot Chat added. The State Data Protection Inspectorate’s guidance from the first pandemic spring still sets the basic rule: a school should choose one set of remote-learning tools rather than leave it to each teacher, and the school, not the teacher, is accountable for them under the GDPR.

The market those schools buy from is small and recently disciplined. The Competition Council found on December 18, 2025 that the two groups behind Lithuania’s main school platforms had agreed in August 2020 to split the market, one keeping the electronic diary business, including TAMO, and the other the digital learning content business, including Eduka klasė, and fined them about €3.6 million in total. Eduka klasė publishes its prices: €19 a year for a student license and €15 for a teacher, with at least seven student licenses per class, ordered centrally by the school. A Lithuanian edtech founder therefore faces a version of the choice this article describes for Baltimore, between selling through schools that buy centrally from a few platforms and selling directly to families.

European law adds its own dates. The AI Act has banned AI systems that infer emotions in education institutions since February 2, 2025, except for medical or safety reasons, which is why the ministry’s guidelines exclude them. Its rules for high-risk AI in education, which cover systems used for admission, grading, assessing the level of education a person receives and proctoring, were due to apply from August 2, 2026, but Regulation (EU) 2026/1744, the Digital Omnibus on AI adopted on July 8, 2026, moved them to December 2, 2027 and softened the duty to ensure staff AI literacy into a duty to take measures to support it. Lithuania’s accessibility law, which carries the European Accessibility Act into national law, has applied since June 28, 2025 to e-commerce services and e-books provided to consumers, with an exemption for microenterprises that provide services and fines of €500 to €2,500 for a non-compliant service. For a family store, the most welcome change is in VAT: since January 1, 2026, printed and electronic books and other non-periodical publications, including textbooks and workbooks, are taxed at 5% instead of 9%, while the standard rate stays at 21%, although we could not confirm whether a PDF set of exercises qualifies as an electronic publication. And under the GDPR as Lithuania applies it, a child of 14 or older may give their own consent to an online service offered directly to them when consent is the legal basis.

The tutoring market that edtalk.lt serves shows what families pay in Lithuania. Across its 54 public tutor profiles in September 2026, an hour of tutoring cost from €15 to €40, with a median of €30, and the first lesson with most tutors costs €10, as its price page sets out. Demand for exam preparation is steady: 14,011 students sat the advanced mathematics matura in 2026 and 85.1% passed, according to the National Agency for Education, and 26,745 sat the tenth-grade mathematics achievement test. Its PDF practice tests and workbooks are the Lithuanian counterpart of the family store described above. For a Lithuanian reader, the lesson of this article is that the American school market asks for far more paperwork before a pilot, an accessibility report, a privacy agreement and now an AI rubric score, and in exchange gives a vendor contracts that run for years across a whole district; a Lithuanian company that wants to sell to American schools should expect to build those records first and the sales pitch second.

What custom actually costs, and when it pays

We price every project at a fixed fee, agreed before we start, and publish the starting points on our pricing page. For a Baltimore edtech company, a tutoring business building its own platform or a curriculum author turning materials into a product, the packages map onto this article like this.

Most of our projects land between $6,000 and $30,000, and none requires you to give up a rostering network, a learning management system integration or a compliance platform that already works; the demos page shows finished work. In Maryland, the state’s 3% tax on software and IT services applies to a custom build, so the Online Store package starts at $6,180 with tax and a Custom App at $12,360.

Whether custom pays depends on what it replaces and what it wins. Set against the first-year district-readiness budget above, a $12,360 district layer is less than half of the typical $26,447 a company will spend anyway on audits, tests and insurance, and it is the part that makes those purchases pass: a penetration test of a product with no tenant isolation fails, and a conformance report for an interface nobody has tested with a screen reader is a liability under §7–910. The family store pays on volume. At our own-store figure of $9.41 kept from a $10 workbook, against $5.20 on a marketplace’s Basic tier, a company keeps $4.21 more per sale and needs about 1,470 sales of that workbook to cover the Online Store package with tax, fewer as soon as the store also books tutoring sessions or sells subscriptions, as the store described above does. And either can pay for itself the first time it turns a “conditional approval” into an approved pilot, or the first time a district’s accessibility reviewer finds a working test account waiting, instead of adding your product to a list like the City’s 59 tools that were still waiting for one.

What we would build for a Baltimore edtech company

We would not build a learning management system, a student information system or a rostering network. Canvas, Schoology, Google Classroom, Infinite Campus, Clever and ClassLink exist so that a district does not have to integrate with every vendor separately, and a three-person company gains nothing by owning its own copy of any of them. We would not build a compliance platform either; the companies that sell SOC 2 automation do that work better and more cheaply than a custom build ever could. What we would build is the layer between the product, the district and the family: the records a principal, a privacy officer, a procurement office, a parent or a regulator can ask about, generated from the company’s own data instead of rebuilt from email the week a contract is due.

The first piece is a data inventory that lives in the code rather than in a spreadsheet. Every field the product stores is tagged with the category of Maryland’s covered information it falls under, whether it is a persistent unique identifier, which feature uses it, which subprocessor sees it and how long it is kept. That one table produces the answers to most of a district’s privacy questionnaire, the data map in a data privacy agreement, the retention schedule the amended COPPA Rule now requires in writing, and the list of what an AI feature is allowed to read. When a developer adds a field without tagging it, the build fails, which is the only way we know to keep a data map true after the first year.

The second is a district desk. Each district is a tenant with its own settings: which schools and grades are enrolled, how students sign in, whether the AI features are on, how long data is kept after the school year ends, and which version of which data privacy agreement governs it, with the date it was signed and the date it expires. A deletion request from a district is a button that deletes the district’s covered information across the database, the file store, the search index, the analytics and the backups as they rotate, and produces a certificate that says what was deleted and when, because “within a reasonable time” is what the statute says and a dated record is what a district will ask for. Every export, every support login to a student’s account and every change to a district’s settings is logged.

The third is an AI switchboard, for any product that uses a language model. It holds, per district and per feature, whether the feature is enabled, which model and provider it calls, what student data is sent and what is stripped first, whether the provider may retain or train on it (for school use the answer should be no, in the contract and in the API settings), and a log a teacher or an administrator can review. It also produces the answers to the questions in Maryland’s AI tool rubric from the same configuration, so the answers stay true when the configuration changes.

The fourth is the family store, built as a separate front door. A parent who buys a plan, a course or a workbook does it on a site that does not share accounts, identifiers or marketing lists with the school product, so that nothing the company learned through a district is ever used to sell to a family, which is the condition Maryland attaches to marketing educational products to parents. The store obtains a parent’s consent before a child under 13 creates an account, charges the right Maryland tax on each line, keeps the receipts and lets a family download or delete what it owns.

The fifth is the boring one that makes the rest believable: an evidence and accessibility record. Usage data, assessment results and implementation notes are kept in a form that a researcher can use for an evidence study later, with the consents and the de-identification that study will need; the accessibility conformance report is regenerated from automated and manual tests whenever the interface changes; and the logic model that a district’s federal grant rules ask for is written down once and kept current. None of it replaces the platforms a district already uses. It connects the product to them through the company’s own data, and all of it is built on your own database, exportable in full, and yours to own.

Build, buy, or leave it alone

Most edtech companies should buy almost everything around their product and build only the product. Single sign-on and rostering should come from Clever, ClassLink or the district’s own identity provider; launches from inside a course should use LTI, the standard every major learning management system already speaks; SOC 2 evidence collection should come from one of the platforms built for it; and payments should go through a processor that already handles cards, bank debits and refunds. The people who build those products have spent years on problems that are invisible until they go wrong, and a founder who rebuilds them is spending the company’s first year on plumbing a district will never thank anyone for.

A teacher who sells worksheets on the side does not need custom software at all, and neither does a tutor who sells a handful of PDFs a month; a marketplace or an off-the-shelf store with a digital download app does that job for a fraction of our price, and we would say so on a call. A founder who has not yet run a pilot should run one with what exists, paper agreements included, before paying anyone to build the district layer, because the pilot is where the company learns which questions its districts actually ask.

Choose your vendors by the meter rather than the logo. Rostering and sign-in networks are priced per school, per app or not at all for the district; certification programs by membership tier and by product; compliance automation by the year and by the framework; audits by the engagement; cyber insurance by the limit the district demands; and payment processors by the transaction. Ask for the full price in writing, including what renews automatically and what the price becomes in year two, and then ask every vendor these questions before you sign:

• Can I see, per district, which students’ data you hold for me, which fields, since when, and delete all of it on one request with a dated record of what was deleted?
• Do you, or your subprocessors, use my students’ data to train or improve any model, and is the answer written into the contract rather than the marketing page?
• Will you sign, or help me sign, the data privacy agreement my districts use, and do you meet the security terms in it, including breach notice and insurance?
• Does your part of my product meet WCAG 2.1 AA, and will you give me an accessibility conformance report I can pass to a district?
• Can I keep my family store’s customers, identifiers and marketing lists completely separate from the data I hold for schools?
• Who owns the data, the code and the integrations, and can I export everything, in full, whenever I like, including after I leave?

A vendor that answers all of them well has earned the contract. Build when the answers are “by hand” for the questions that carry your risk: when the data map lives in a founder’s memory, when a district’s deletion request is a support ticket that someone closes without a record, when nobody can say which version of which agreement a district signed, when the AI feature sends whatever a student types to whichever model was cheapest that month, or when the store that sells to parents shares a login table with the product the schools use. Leave alone what is cheap and works: the learning management systems, the rostering networks, the identity providers and the compliance platforms your districts already trust.

Our guide to custom software in Baltimore explains how we work, our article on tutoring and test prep covers the background checks and contracts of companies that send people into schools, our article on homeschool co‑ops and microschools covers the families who are not in a district at all, and our article on Maryland’s Online Data Privacy Act covers the law that applies to the family store. If you want to find out which of these would pay for you, bring us one district’s questionnaire, your data privacy agreement and your current data map, and we will show you.

Frequently asked questions

Does Maryland have a student data privacy law?

Yes. The Student Data Privacy Act of 2015, Education Article §4–131, rewritten by Chapters 163 and 164 of 2022, applies to an “operator”: a company that works with a Maryland public school or school system under FERPA’s school official exception and a contract or agreement, and whose website, online service or app processes covered information for a PreK–12 school purpose or is issued at the direction of a school, a teacher or another employee. It must protect covered information, maintain reasonable security and delete a school’s data within a reasonable time on request, and it may not use the data for targeted advertising, build non-school profiles of students, sell student information or disclose it except in listed situations. It does not apply to general audience products that families find and buy on their own. Maryland’s Online Data Privacy Act and Kids Code can apply to those instead.

What counts as covered information under Maryland’s Student Data Privacy Act?

Covered information is information that, alone or combined with other information, could let someone at a student’s school identify the student with reasonable certainty. The statute lists 23 kinds, including educational records, names, addresses and geolocation, grades and test results, special education, health, disciplinary and biometric information, text messages, search activity, photos, voice recordings, online behavior linked to a student and persistent unique identifiers. That last category includes cookie identifiers, customer numbers, device identifiers, hashed email addresses and phone numbers, probabilistic identifiers, mobile ad identifiers, unique pseudonyms and user aliases, so a hashed email or a classroom nickname is covered information.

How do I become a vendor for Baltimore City Public Schools?

Register on eMaryland Marketplace Advantage (eMMA) to receive bid notices and on the district’s Oracle supplier portal to receive purchase orders and submit invoices. City Schools buys in a decentralized way: a school or office can buy up to $25,000 with one quote, more than $25,000 up to $50,000 with two and more than $50,000 up to $100,000 with three, and a vendor without Board approval may not be paid more than $100,000 in a fiscal year across the district. Anything larger goes to the Board of School Commissioners, usually after an invitation for bids or a request for proposals. Any purchase involving digital tools or data sharing needs a written agreement rather than just a purchase order, and only the chief executive officer and the Director of Procurement can sign contracts.

How do I become a vendor for Baltimore County Public Schools?

Baltimore County’s Supplier Guide, revised in September 2026, asks vendors to register in its Oracle supplier system with a W-9, good standing with the State Department of Assessments and Taxation and the commodity codes they sell under. Purchases under $15,000 need written quotes, purchases from $15,000 to under $100,000 need at least two quotes, and purchases of $100,000 or more need an invitation for bids or a request for proposals and Board approval. Instructional tools are reviewed by a curriculum committee before purchase, and the district’s software contracts include its student data privacy requirements. Many purchases are made through cooperative contracts: 78 of the 506 contracts in the district’s August 2026 database were cooperative agreements.

Can I run a free pilot in a Baltimore City school?

Yes, but not on a handshake. In City Schools an arrangement in which the district pays nothing is handled as a memorandum of understanding, which needs the approval of the responsible chief officer, the district’s legal counsel, the Director of Procurement and the chief executive officer or the chief of staff, and any tool that touches student data needs a written agreement whatever it costs. If the tool uses AI, Maryland’s rubric asks four questions before any pilot, starting with a contractual promise not to train foundational models on district data.

Do edtech companies need a VPAT to sell to Maryland schools?

In practice, yes. Since September 1, 2023, Education Article §7–910 has required every invitation for bids or request for proposals for a digital tool issued by a Maryland school system to require an accessibility conformance report that includes a Voluntary Product Accessibility Template and explains conformance to the most recent Section 508 standards. The school system must also have the tool evaluated by an accessibility or blindness specialist, its contract must require the vendor to indemnify it, and a vendor that does not fix a failing tool after notice faces civil penalties of up to $5,000 for a first offense and $10,000 for a later one. Because the statute refers to Section 508, the Section 508 or combined international edition of the VPAT is the one to prepare.

Are AI tools exempt from Maryland’s accessibility requirements for schools?

Temporarily, from Maryland’s own statute. Chapter 237 of 2025 added a subsection to Education Article §7–910 under which, from the 2025–2026 school year through the 2027–2028 school year, the section does not apply to the procurement and use of a digital tool that uses artificial intelligence to support student learning; the change lapses at the end of June 30, 2028. Each school system decides whether a tool uses AI, and the fiscal 2026 reports due in October 2026 add a list of the AI tools each district bought. The exemption does not cover the federal ADA rule, which requires WCAG 2.1 AA for content and apps school districts provide, including through contracts, from April 26, 2027 for both Baltimore systems.

What is Maryland’s AI rubric for schools?

Chapter 634 of 2026, the Artificial Intelligence Ready Schools Act, took effect on June 1, 2026 and requires each local school system to use the State Department of Education’s rubric when reviewing and selecting AI tools, to adopt an AI policy within 120 days of the state’s guidance and to name an AI coordinator. The rubric presented to the State Board in February 2026 scores tools on nine criteria, from data privacy and human oversight to LMS interoperability and environmental impact. Before a pilot, four answers must be yes: a contractual promise not to use district data to train foundational models, labeled AI content, a visible way to flag errors or inappropriate content, and a process for permanent deletion on request. A score of 24 to 28 means approved for a pilot, 18 to 23 conditional approval and below 18 denied.

When do school district websites and apps have to meet WCAG 2.1 under the ADA?

Under the Justice Department’s Title II rule, as extended by an interim final rule published on April 20, 2026, public entities with a population of 50,000 or more must comply by April 26, 2027 and smaller ones and special district governments by April 26, 2028. Both Baltimore school systems are in the first group. The rule covers web content and mobile apps a public entity provides “directly or through contractual, licensing, or other arrangements,” which includes third-party edtech, and it has no exception for course content behind a password. Maryland’s own statute asks for “the most recent version” of WCAG, which today is 2.2.

Does COPPA apply to edtech used in schools?

Yes, when a product collects personal information from children under 13. The FTC lets a school consent on parents’ behalf only where the data is collected for the use and benefit of the school and for no other commercial purpose, and the company must give the school notice of its practices and let it review and delete the data; the FTC declined to write that school authorization into the amended rule in 2025 and continues to rely on its existing guidance. The amended COPPA Rule, with which companies had to comply by April 22, 2026, also requires a written information security program, a written data retention policy in the privacy notice and a separate parental consent for disclosures that are not integral to the service, which the FTC says includes disclosures to train AI.

Can an edtech company market to parents in Maryland?

Yes, with one condition. Education Article §4–131(k) says the Student Data Privacy Act may not be read to stop an operator from marketing educational products directly to parents “if the marketing was not a result of the use of covered information” obtained through the school service. In practice that means separate customer records, identifiers and marketing lists for the family side, and no recommendations or promotions inside the school product based on what the company learned there. Maryland’s Online Data Privacy Act also bars selling the data of, or targeting ads at, anyone a covered company knew or should have known was under 18.

Is edtech software taxable in Maryland?

It depends on the buyer. Sales to a public school system are exempt as sales to government, and sales to a Maryland nonprofit charitable or educational organization with the Comptroller’s exemption certificate are exempt too. A business buying software as a service for its own enterprise systems pays 3% under the tax on data and IT services that began on July 1, 2025, and a family buying the same subscription for personal use pays 6%, because it is a digital product and the higher rate applies. Downloadable workbooks and recorded courses are digital products at 6%, while live tutoring generally is not. This is our reading for the businesses we build for, not tax advice.

What is ESSA Tier 4 evidence, and how does a product get rated by Evidence for ESSA?

Tier 4, “demonstrates a rationale,” is the entry level of the Every Student Succeeds Act’s evidence tiers: a research-based logic model plus ongoing efforts to examine the product’s effects. Tiers 1 to 3 require experimental, quasi-experimental or correlational studies, and school improvement money under section 1003 of Title I can only buy programs in those tiers. Evidence for ESSA, run by Johns Hopkins University’s Center for Research and Reform in Education in Baltimore, rates programs in the top three tiers only; its June 2026 standards require at least two schools and 350 students for a strong or moderate rating, accept no developer-made measures and only American studies of at least twelve weeks, and invite developers to submit studies by email. On October 17, 2026 it listed 81 programs as strong, 38 as moderate and 77 as promising out of 1,292.

How much does it cost to make an edtech product district-ready?

At prices published on October 17, 2026, a three-person company with one web app should budget roughly $9,900 to $68,400 in its first year: SOC 2 from a $3,555 bundled Type I report to about $37,500 for a platform and a Type 2 audit, a penetration test from $2,999 to $10,800, cyber insurance from a median $1,552 a year, though districts can require $2 million of coverage, an accessibility audit with a Section 508 conformance report from about $1,800, and optional seals such as ISTE’s at $5,000 plus $1,500 a year. That excludes the engineering work and several prices nobody publishes, including Clever’s fee to app partners and 1EdTech membership dues.

What changed for Teachers Pay Teachers sellers in October 2026?

On October 1, 2026 Teachers Pay Teachers closed its Premium seller membership to new sign-ups and introduced an Exclusive membership at $59.95 a year that pays 80% of each sale until January 1, 2027 and 77.5% after that, with a 15-cent fee per resource only on orders under $3. Exclusive sellers must sell their digital educational resources, memberships that include them and licenses for them only on Teachers Pay Teachers, so their own websites can promote those resources but must send buyers to the marketplace. The Basic membership, $29 once, still pays 55% minus 30 cents per resource and allows selling elsewhere. Existing Premium sellers keep 80% through December 31, 2026 and must choose by then.

How much does it cost to build an education app?

Our fixed prices start at $3,500 for a one-week Prototype Sprint, $6,000 for an online store and $12,000 for a custom app or an operations system, and most of our projects land between $6,000 and $30,000; Maryland’s 3% tax on software services is added on top. For an edtech company the larger costs are usually not the first version of the product but the district layer around it, rostering, LTI, deletion, logs, accessibility and AI controls, and the audits and insurance described above.