The Student Data Frontier: How 100 State Bills, a Federal Literacy Act, and a Deepfake Crisis Are Rewriting AI Rules for K-12 Classrooms
Nearly 100 state bills, a federal AI literacy act, and soaring deepfake bullying incidents converge to reshape what AI can and cannot do in American schools.
The Scale of the 2026 Legislative Wave
Artificial intelligence has moved rapidly into K-12 classrooms, from tutoring tools to content generation platforms, and state policymakers are scrambling to keep up. In 2026, the PIE Network tracked nearly 100 state bills that could directly affect students’ use of AI in K-12 education, part of more than 1,500 AI-related bills introduced by lawmakers nationwide this year[1]. MultiState, a government relations tracking firm, counted 134 bills across 31 states focused specifically on AI in education[2]. That activity signals that states have moved beyond early exploration of AI’s capabilities and into active governance, grappling with a core tension: how to embrace the promise of educational innovation while establishing clear guardrails to protect students and preserve trust in public education[1].
Three policy themes dominate the 2026 wave: student data privacy protections, AI usage boundaries and human oversight requirements, and AI literacy and graduation requirements. Each addresses a different facet of the same question: when AI systems interact with children in educational settings, who controls the data, who oversees the decisions, and who ensures students understand the technology shaping their learning?
The Data Privacy Battleground
Protecting student information has emerged as the top priority for state lawmakers. The central concern is that educational technology vendors could collect student data through school-sanctioned tools and then use that data to train commercial AI models, effectively turning children’s learning records into raw material for products unrelated to their education.
California AB 1159: The Broadest Expansion
California’s A.B. 1159, introduced by Assemblymember Dawn Addis and currently under Senate consideration, represents the most comprehensive state-level attempt to close the gap between existing student privacy law and the realities of generative AI[3]. The bill would prohibit operators of internet websites, online services, and applications used for school purposes from using “covered information,” including persistent unique identifiers, to train generative AI systems[4]. It also broadens the definition of “operator” to cover any entity with actual knowledge that its service is used for educational purposes and was designed or marketed for those purposes, closing a loophole that allowed companies to claim ignorance about how their tools were deployed in schools[4].
The bill goes further than data training prohibitions in several important ways:
- Higher education expansion: AB 1159 enacts the Higher Education Student Information Protection Act (HESIPA), extending privacy protections to college students, who currently lack the K-12 safeguards that laws like KOPIPA and ELPIPA provide[4]. This bridges an existing regulatory gap, as protections have historically stopped at high school graduation.
- Sensitive data carve-outs: The bill prohibits the collection, use, retention, or disclosure of information relating to a student’s reproductive or sexual health, immigration status, sexual orientation, or gender identity[3]. Assemblymember Addis framed these protections as necessary given that the federal government “has tried to collect increasing amounts of personal information”[3].
- Private right of action: AB 1159 authorizes a pupil or student actually harmed by noncompliance to bring a civil action against the noncompliant operator[4]. This enforcement mechanism is notable because most student privacy laws rely exclusively on state attorneys general or education agencies to pursue violations, leaving affected families with little direct recourse.
The bill has drawn organized opposition. TechNet and the College Board argue it is overly broad and would “chill” responsible AI development in education technology[3]. The College Board specifically contends that the bill would fail to permit essential disclosures needed for students to use test scores for scholarships, college placement, and course credit[3]. Supporters, including the Privacy Rights Clearinghouse, counter that AB 1159 does not prohibit the use of de-identified personal information for AI product development, undermining the industry argument[3].
Idaho SB 1227: The Enacted Framework
While California debates, Idaho has already acted. Idaho’s S.B. 1227, now enacted, requires a statewide framework for AI in K-12 schools, mandates local policies for its use, establishes AI literacy standards and educator training, sets data privacy requirements for AI tools, and prohibits AI from replacing human teachers[2]. The bill is significant because it treats AI in education as a system requiring comprehensive governance rather than a single problem to be regulated in isolation. By pairing privacy requirements with literacy standards and educator training mandates, Idaho’s approach recognizes that data protection without capacity-building leaves schools unable to meaningfully evaluate the tools they adopt.
The FERPA Enforcement Gap
Underlying these state efforts is a federal law that many assume already protects student data: the Family Educational Rights and Privacy Act (FERPA). FERPA governs how schools share student education records with vendors through the “school official exception,” which permits disclosure without parental consent when an outside party performs a service for which the school would otherwise use its own employees, the school maintains direct control over the data, and the outside party uses it only for the authorized purpose[5]. The regulations at 34 CFR Section 99.33 limit redisclosure: once a vendor receives FERPA-protected data, it cannot share it with subprocessors unless they agree to the same restrictions, and it cannot use the data for any purpose beyond what the school authorized[5].
The problem is that FERPA was written for an era of filing cabinets and mail, not machine learning pipelines. If an AI model is trained on student records shared under the school official exception, and that training persists in the model’s weights after the contract ends, the vendor has not truly “deleted” the data; it is baked into the model[5]. Experts warn that FERPA still lacks clear cybersecurity requirements even though schools rely on hundreds of ed-tech tools, and the Department of Education has doubled down on enforcement and transparency in 2025, requiring schools and state agencies to demonstrate proactive protections rather than reactive responses[6]. But without explicit rules addressing model training and weight retention, a gap remains that state bills like California’s AB 1159 and Idaho’s SB 1227 are attempting to fill.
The Federal Push: K-12 AI Literacy and Readiness Act of 2026
At the federal level, H.R. 8747, the K-12 AI Literacy and Readiness Act of 2026, was introduced by Rep. Randy Fine (R-FL) on May 12, 2026, and ordered to be reported by the House Education and Workforce Committee on July 21, 2026, by an 18-15 vote[7]. The bill amends the Elementary and Secondary Education Act of 1965 to include AI-related curriculum expenses as permissible uses of federal Title IV-A funds[8].
Specifically, the bill adds two categories of allowable spending:
- Student instruction: Funds may be used for “the use of artificial intelligence in a manner that is safe, effective, and responsible”[8]. This language is deliberately broad, giving districts flexibility to design age-appropriate AI literacy curricula rather than mandating a specific program.
- Teacher and school leader training: Funds may be used for “providing teachers, paraprofessionals, school librarians and media personnel, specialized instructional support personnel, and administrators with the knowledge and skills to use, and teach the use of, artificial intelligence safely, effectively, and responsibly, including through professional development”[8]. The inclusion of librarians and media personnel is notable, reflecting the reality that AI literacy often begins in school library and media center settings.
The bill’s approach is incremental rather than prescriptive. It does not mandate AI instruction or create new regulatory requirements. Instead, it makes AI literacy an allowable use of existing federal funding streams, lowering the barrier for districts that want to invest in AI education but have been constrained by rigid Title IV-A spending rules. The legislation defines AI by cross-referencing section 5002 of the National Artificial Intelligence Initiative Act of 2020, anchoring the term in existing federal statute rather than crafting a new definition that could quickly become outdated[8].
The Deepfake Crisis in Schools
While legislatures debate data privacy and literacy standards, a more immediate harm is already reshaping the school safety landscape: AI-generated deepfakes used for cyberbullying. A RAND Corporation survey of a nationally representative sample of K-12 school principals, fielded in October 2024, found that 13 percent reported incidents of bullying involving AI-generated deepfakes during the 2023-2024 and 2024-2025 school years[9]. The problem is concentrated in secondary schools: 22 percent of high school principals and 20 percent of middle school principals reported such cases, compared with 8 percent of elementary school principals[9]. For these older student populations, roughly one in five schools has dealt with bullying and victimization using AI-generated images or video[9].
A 2024 survey by the Center for Democracy and Technology found even higher rates of student exposure: 40 percent of K-12 public school students were aware of a deepfake involving someone at their school, including 15 percent who knew of a sexually explicit deepfake depicting a member of their school community[10]. More than 60 percent of teachers reported their school had not shared policies or procedures related to AI-generated sexual imagery[10], suggesting that most schools were unprepared for the scale of the problem.
Illinois: The First State to Name Deepfakes as Cyberbullying
Illinois has become the first state to explicitly amend its school cyberbullying definition to include AI-generated digital replicas. Beginning July 1, 2026, the state’s definition of cyberbullying was expanded to include AI-generated content, forcing districts across Illinois to update their policies and procedures[10]. The Illinois State Board of Education was also required to develop statewide guidance on artificial intelligence in schools, expected to address AI-generated false representations of individuals to help districts update their policies ahead of the 2026-27 school year[10].
The law does not change Illinois’ criminal statutes governing AI-generated sexual imagery. Under state law, AI-generated sexual images that appear to depict minors may be prosecuted under child pornography and related obscenity statutes, with felony charges possible for creating, possessing, or distributing such images[10]. Scott Rowe, superintendent of Township High School District 214, framed the legislation’s core principle succinctly: “AI does not remove responsibility”[10].
How schools have responded to incidents, according to the RAND survey, reveals a gap between reaction and prevention. Among schools that experienced deepfake incidents:
- 79 percent took disciplinary actions against those involved[9]
- 66 percent referred incidents to law enforcement[9]
- 47 percent provided education and training to staff and students on recognizing deepfakes[9]
- Only 23 percent updated school or district-wide policy to include specific clauses about AI misuse[9]
The low rate of policy updates is striking. Most schools are responding to deepfake incidents reactively, through discipline and law enforcement referrals, rather than proactively updating their policies to define AI misuse and establish prevention protocols. Kristin Woelfel, a policy counsel for the Center for Democracy and Technology, argues that updated policies must go beyond consequences: they also need to establish reporting protocols and support structures for victims. As she put it, “It’s not just like ‘You’ll get in trouble if you do this.’ It should also be like, ‘Here is who you report something like this to.’”[10]
Massachusetts: State Guidance for Schools
In April 2026, Governor Maura Healey and the Massachusetts Department of Elementary and Secondary Education issued formal guidance confirming that deepfakes are covered under existing law and that schools must investigate complaints promptly, consistent with Title IX, bullying, and harassment obligations[11]. The guidance reinforces that creating, possessing, or sharing AI-generated explicit images of a minor is not just a violation of school rules but a criminal offense under Massachusetts law, with penalties of up to 2.5 years in prison and a $10,000 fine[11]. Pending legislation in the state, H.1593, would expand existing laws to explicitly include AI-generated child sexual abuse material[11].
AI Literacy as Graduation Requirement
Beyond data privacy and deepfake response, a growing number of states are embedding AI literacy directly into academic standards and graduation requirements. This trend reflects a shift in how policymakers view AI: not merely as a risk to be managed, but as a core competency students need to navigate an increasingly AI-connected world.
- Utah enacted legislation adding AI literacy to middle school academic standards[1]. Core logic: Introducing AI literacy before high school ensures students develop critical judgment about the technology before they encounter it in higher-stakes academic and professional contexts.
- Mississippi passed S.B. 2294, requiring high school students starting with the 2029-2030 ninth-grade class to earn a computer science or CTE credit that includes instruction on emerging technologies such as AI[2]. Core logic: Tying AI literacy to a graduation requirement rather than an elective ensures all students, not just those in specialized tracks, gain exposure.
- Georgia passed S.B. 179, making computer science, including AI, a high school graduation requirement beginning in 2031-2032 and phasing in statewide access to computer science instruction[2]. Core logic: The phased approach acknowledges that rural and under-resourced districts need time to build capacity before a mandate takes effect.
These literacy requirements represent a different philosophical posture from the data privacy and deepfake bills. Where privacy legislation treats AI as a threat to be contained, literacy standards treat it as a skill to be cultivated. The strongest policy frameworks, as ExcelinEd notes, will do both: protect students from harm while equipping them to engage critically with AI systems[1].
Usage Boundaries: Keeping Humans in the Loop
A third strand of 2026 legislation focuses on what AI cannot do in educational settings, particularly around high-stakes decisions. Oklahoma’s S.B. 1734, which passed its first chamber, allows AI in schools only under educator supervision with human review, bans its use for high-stakes decisions, mandates state guidance and district policies, and requires annual parent disclosure[2]. Maryland’s S.B. 720 and H.B. 1057 require the State Department of Education to issue AI guidance, mandate local school systems to adopt AI policies and promote AI literacy, and establish a statewide collaborative to study AI in K-12 education[2]. New York’s A. 9190 would restrict AI use in classrooms to ninth grade and above, except for diagnostics or special education interventions[2].
The common principle across these bills is that AI should function as a support tool, not a decision-maker. The underlying logic is straightforward: AI systems can process vast amounts of student data and generate recommendations, but they lack the contextual judgment, empathy, and professional expertise that human educators bring to high-stakes decisions about student placement, grading, discipline, and special education services. By requiring human oversight, these bills codify a boundary that many educators would consider common sense but that, without legal protection, could erode under budget pressures and vendor marketing.
What Is Still Missing
Despite the volume of legislative activity, significant gaps remain in the emerging AI policy framework for K-12 students.
The District Vetting Burden
Responsibility for evaluating AI tools falls heavily on local districts, many of which lack the technical and legal capacity to assess complex AI systems. States could address this by providing standardized evaluation frameworks, vetted procurement criteria, or centralized libraries of approved tools, but few have done so[1]. The result is fragmented decision-making and uneven student protections across localities, with well-resourced districts conducting thorough vendor reviews while under-resourced districts rely on vendor marketing claims.
Vendor Transparency
Most current proposals do not require vendors to provide auditable records of user interactions, disclose how systems generate outputs, or demonstrate that tools have been evaluated for bias, accuracy, and reliability before being used by students[1]. Without these transparency requirements, school leaders approach procurement decisions with incomplete information. A district purchasing an AI tutoring system may have no way to know whether the system’s training data reflects the demographics of its student population, whether its recommendations have been tested for bias, or whether student interactions are logged and reviewable.
The FERPA Modernization Gap
While states are filling gaps left by FERPA’s age, the federal law itself has not been meaningfully updated to address AI-specific challenges. Experts urge lawmakers to modernize FERPA with vendor security obligations and explicit rules about model training and data retention in AI weights[6]. Until that happens, the patchwork of state laws will produce inconsistent protections: a student in California may have stronger data privacy rights than a student in a state with no AI-specific legislation, simply because of geography.
Conclusion
The 2026 legislative wave marks a turning point for AI policy in K-12 education. States have moved from asking whether AI belongs in classrooms to asking how to govern it. California’s AB 1159 and Idaho’s SB 1227 represent two models for protecting student data: one through comprehensive statutory expansion with a private right of action, the other through an enacted framework that pairs privacy with literacy and training mandates. The federal K-12 AI Literacy and Readiness Act, if passed, would lower the funding barrier for districts seeking to build AI literacy programs. And the deepfake crisis, documented in RAND’s 13 percent principal-reported incident rate, has forced states like Illinois and Massachusetts to confront a harm that existing cyberbullying frameworks were not designed to address.
The challenge for the next phase of AI education policy is moving from isolated requirements to coherent implementation. Privacy protections without capacity-building leave schools unable to evaluate the tools they adopt. Literacy standards without vendor transparency leave students learning to critique systems whose inner workings remain opaque. And reactive deepfake response without proactive policy leaves schools perpetually one step behind the technology. The strongest frameworks will do what the best of the 2026 bills already attempt: protect students from harm, equip educators with the knowledge to guide responsible use, and ensure that AI serves the educational mission rather than the other way around[1].
References
[1] [SECONDARY] Song, Ji Soo. “State K-12 AI Policy in 2026: Milestones, Momentum and Missing Links.” ExcelinEd. May 26, 2026. https://excelined.org/2026/05/26/state-k-12-ai-policy-in-2026-milestones/
[2] [SECONDARY] Aaron, Izzy. “How States Are Regulating AI in Education this Legislative Session.” MultiState. April 9, 2026. https://www.multistate.us/insider/2026/4/9/how-states-are-regulating-ai-in-education-this-legislative-session
[3] [SECONDARY] Khawaja, Aisha S. “AB 1159: Extending Student Data Privacy Protections to Higher Education.” Medium / Golden Data. June 9, 2026. https://medium.com/golden-data/ab-1159-extending-student-data-privacy-protections-to-higher-education-8676161660e1
[4] [PRIMARY] California Legislature. “AB 1159: Student Personal Information.” 2025-2026 Session. Amended August 3, 2026. https://calmatters.digitaldemocracy.org/bills/ca_202520260ab1159
[5] [SECONDARY] Promise Legal. “EdTech AI Compliance: FERPA, COPPA & State Laws.” blog.promise.legal. 2026. https://blog.promise.legal/edtech-ai-compliance-ferpa-coppa/
[6] [SECONDARY] Concentric AI. “A Guide to Using AI for FERPA Compliance.” concentric.ai. 2026. https://concentric.ai/maintain-ferpa-compliance-with-concentric-ai/
[7] [PRIMARY] U.S. Congress. “H.R.8747 – K-12 AI Literacy and Readiness Act of 2026, All Actions.” 119th Congress. https://www.congress.gov/bill/119th-congress/house-bill/8747
[8] [PRIMARY] U.S. Congress. “H.R.8747 – K-12 AI Literacy and Readiness Act of 2026, Bill Text.” 119th Congress. Introduced May 12, 2026. https://www.congress.gov/bill/119th-congress/house-bill/8747/text
[9] [SECONDARY] Jackson, Brian A., Melissa Kay Diliberti, and Pauline Moore. “Artificially Intelligent Bullies: Dealing with Deepfakes in K-12 Schools.” RAND Corporation, RR-A3930-5. September 2025. https://www.rand.org/pubs/research_reports/RRA3930-5.html
[10] [SECONDARY] Taxay, Naomi. “Illinois Schools Grapple with AI Cyberbullying, Deepfakes as New Law Takes Effect.” Capitol News Illinois. July 3, 2026. https://capitolnewsillinois.com/news/illinois-schools-grapple-with-ai-cyberbullying-deepfakes-as-new-law-takes-effect/
[11] [SECONDARY] Holliston Public Schools. “AI Deepfake Guidelines.” Massachusetts DESE Guidance. April 2026. https://hhs.holliston.k12.ma.us/ai-deepfake-guidelines
AI automated and generated content. Human reviewed.
The Weekly AI Policy Brief
AI policy moves fast. Get the week’s developments, new trackers, and podcast episodes in one short email — free, ad-free, no spam.
Leave a Reply