New Mexico's Public Education Department has proposed rule 6.11.3 NMAC, which would set student data privacy requirements for contracts between K-12 schools and education technology vendors. The rule is scheduled to take effect April 1, 2027, and its public comment period runs through November 20, 2026, the department stated in its notice of proposed rulemaking. Public hearings are set for November 19 and 20.
Rule allows AI training with contractual authorization
The department said in written responses to questions from Organ Mountain News that the proposed rule does not categorically prohibit using student data to train, develop, or improve AI tools that support a contracted educational service. Such use must be authorized in the contract and comply with other restrictions. A vendor cannot justify using student information simply by calling it service improvement, the department added.
The proposed rule text outlines several obligations that vendor contracts must include. Vendors must restrict data access to authorized personnel with a legitimate need. They cannot use student data for targeted advertising or create student profiles except as necessary for the contracted service or with prior written approval. Selling, leasing, or trading student data to third parties is prohibited except as needed for the contracted service. Vendors must destroy student data within 30 days of a written request or contract termination, implement reasonable security procedures, and provide at least 60 days written notice before any material change to privacy policies or data-handling practices.
The rule also requires breach notification. Vendors must notify the contracting district or the department within 48 hours of discovering an unauthorized disclosure, and districts must notify parents within 72 hours of receiving that notice, according to the rule text. Before a contract is signed, vendors must provide a plain-language data privacy impact assessment and a security assessment, which may be satisfied by an independent third-party audit or a recognized framework such as a SOC 2 Type II report from a named auditing firm.
The rule defines student data broadly to include personally identifiable information from education records, information provided by students or parents to vendors for K-12 school purposes, and information gathered by vendors that identifies a student. The definition covers test results, health records, biometric information, disability status, socioeconomic information, and persistent unique identifiers, according to the proposed rule text.
On biometrics, the rule defines voiceprint as an audio recording processed to create a unique vocal signature capable of identifying a specific student. Recordings that are not processed into a persistent identifier are excluded. Schools must give parents written, standalone notice before collecting biometric information, explaining what will be collected, its purpose, storage, retention period, and destruction method. Affirmative parental consent for biometric collection is required only when otherwise required by law. The department said a universal opt-in requirement could prevent students from participating because a parent did not return a signed consent form, and some schools may lack resources for alternative instructional services.
Contrast with California and Santa Fe
California took a different approach in September 2026 when Governor Gavin Newsom signed AB 1159 into law, according to the bill text. That law explicitly prohibits operators from using covered student information to train a generative AI system or service. It applies to K-12, early learning, and higher education and creates a private right of action for harmed students. CalMatters reported that the law expands upon California's existing Student Online Personal Information Protection Act, which New Mexico's PED cited as a framework for its proposed rule.
Santa Fe Public Schools moved in the opposite direction from the state rule. According to the Santa Fe New Mexican and GovTech, the district's school board considered a draft student data privacy policy, Policy 357, that would bar vendors from using student data to train, develop, or improve artificial intelligence models. The board delayed a vote on Policy 357 while separately approving a cybersecurity policy, Policy 547. The draft student data privacy policy would require affirmative written opt-in consent from a parent before using sensitive data for automated analysis, and provide an equivalent alternative without penalty. This policy is more restrictive than the state-level rule.
Debate over AI training on student data
The question of whether student data should be used to train AI models remains contested. The Privacy Rights Clearinghouse, which sponsored California's AB 1159, argued that AI training on student data is irreversible. Once a company feeds student information into a model, the data cannot be taken back, making prohibition justified, according to the group. The College Board, during California's legislative proceedings, argued that a ban would effectively bar students and educators from accessing educational tools that responsibly incorporate AI features, as noted in Formative Spaces' coverage of the California law. The California Assembly Privacy and Consumer Protection Committee analysis countered that the bill does not prohibit use of deidentified information for AI training, and questioned why that limitation would bar such tools. No peer-reviewed research was identified evaluating whether AI training bans or restrictions improve student data privacy outcomes.
Context: Amira reading assessment
The proposed rule follows months of concern about student voice recordings collected through the Amira reading assessment. In August 2026, 35 New Mexico lawmakers urged PED to strengthen parental consent and oversight surrounding the assessment, according to Organ Mountain News. Las Cruces Public Schools subsequently disabled retention of student voice recordings and obtained written confirmation that previously stored recordings had been deleted. The district made a paper-based assessment available to parents who did not want their children tested digitally. PED stated that the Amira program does not process voice recordings into voiceprints for biometric identification, but recordings linked to a student's name or account may still constitute protected student data, according to the rule text.
The proposed rule's definition of voiceprint and its carve-out for recordings not processed into a persistent identifier drew criticism in public comments. One commenter urged PED to remove the carve-out, saying it appeared written with Amira in mind and could allow a vendor to continue recording students and using the recordings for model improvement, according to the department's compilation of public comment exhibits.