SECTION 164 of the Senate-engrossed Massachusetts economic development bill inserts two new chapters into the General Laws after chapter 93L: chapter 93M, the Transparency in Frontier Artificial Intelligence Act, and chapter 93N, governing commercial use of chatbots. The vehicle is H.5576, An Act relative to economic development in the commonwealth; the operative text is the Senate amendment S.3228, which struck everything after the enacting clause and substituted the text of S.3178, reprinted as amended and engrossed July 24, 2026. The AI language sits before a committee of conference and is not law.
Definitions
| Term | Test |
|---|---|
| Foundation model | An AI model trained on a broad data set, designed for generality of output, and adaptable to a wide range of distinctive tasks |
| Frontier model | A foundation model trained using more than 10^26 integer or floating-point operations, counting the original training run plus any subsequent fine-tuning, reinforcement learning, or other material modifications the developer applies to a preceding foundation model |
| Frontier developer | A person who has trained, or initiated the training of, a frontier model, having used or intending to use at least as much computing power as the frontier-model specification requires |
| Large frontier developer | A frontier developer that, together with its affiliates, has annual gross revenues greater than $500,000,000 |
| Covered employee | An employee responsible for assessing, managing or addressing risk of critical safety incidents |
Catastrophic risk is "a foreseeable and material risk" that a frontier developer's development, storage, use or deployment of a frontier model "will materially contribute to the death of, or serious injury to, not less than 50 people or not less than $1,000,000,000 in damage to, or loss of, property arising from a single incident" involving a frontier model that either (i) provides expert-level assistance in creating or releasing a chemical, biological, radiological or nuclear weapon; (ii) engages in conduct with no meaningful human oversight that is a cyberattack or that, if committed by a human, would constitute murder, assault, extortion or theft; or (iii) evades the control of its developer or user. Three carve-outs apply: information otherwise publicly accessible in substantially similar form from a non-foundation-model source; lawful federal government activity; and harm caused in combination with other software where the frontier model did not materially contribute. A separate section provides that "[t]he loss of value of equity shall not constitute damage to or loss of property" for the chapter's purposes.
Critical safety incident covers four events: unauthorized access to, modification of, inadvertent release of, or exfiltration of a frontier model's weights; harm from the materialization of a catastrophic risk; loss of control causing death or bodily injury, or demonstrating materially increased catastrophic risk; and an instance where a frontier model "uses deceptive techniques against the frontier developer to subvert the controls or monitoring of its frontier developer," outside an evaluation designed to elicit that behavior, in a manner demonstrating materially increased catastrophic risk.
Publication obligations
A large frontier developer must write, implement, comply with and publish a frontier AI framework addressing eleven enumerated matters: incorporating national, international and industry-consensus standards; defining and assessing capability thresholds for catastrophic risk, which "may include multiple-tiered thresholds"; applying mitigations; assessing the models' ability to automate AI research and development and any resulting increase in catastrophic risk; reviewing assessments as part of the deployment decision or extensive internal use; using third parties to assess risk and mitigation effectiveness; revisiting and updating the framework; cybersecurity practices securing unreleased model weights; identifying and responding to critical safety incidents; internal governance; and managing catastrophic risk from internal use, "including risks resulting from a frontier model circumventing oversight mechanisms." The framework is reviewed at least annually, and a material modification must be published with a justification within 30 days.
Every frontier developer must publish a transparency report before or concurrently with deploying a new or substantially modified frontier model, covering the developer's website address, a means of contacting a natural person, the release date, supported languages and output modalities, intended uses, and any generally applicable use restrictions. A large frontier developer must add summaries of its catastrophic-risk assessments, their results, the extent of third-party involvement, and other framework-compliance steps. Publication inside a larger document such as a system card or model card satisfies the requirement.
A large frontier developer must also publish a risk report covering externally deployed frontier models and any internally deployed model whose capabilities materially exceed those of anything it has deployed externally. The report must summarize capability assessments relevant to each type of catastrophic risk and material changes since the last report, describe the key threat models tracked and the mitigations in place, and assess residual risk with "sufficient information to demonstrate the evidence and reasoning behind the risk assessment" — enough "to allow a reasonable person to reach a similar conclusion." Risk reports are renewed at least every 180 days, each update comparing assessed risk levels to the previous report. Separately, a summary of any catastrophic-risk assessment arising from internal use goes to the Attorney General every 3 months, or on another schedule agreed with the Attorney General.
Two anti-misstatement provisions bar materially false or misleading statements about catastrophic risk and its management (all frontier developers) and about implementation of or compliance with the frontier AI framework (large frontier developers), subject to a good-faith-and-reasonable exception. Redactions are permitted for trade secrets, cybersecurity, public safety, national security or legal compliance, but the character and justification of each redaction must be described and the unredacted information retained for five years.
Incident reporting
The Attorney General must establish a mechanism through which a frontier developer or a member of the public can report a critical safety incident, and a separate confidential mechanism for large frontier developers to submit internal-use risk assessments, with access limited to authorized personnel.
A frontier developer must report a critical safety incident to the Attorney General within 15 days of discovery; where the incident poses an imminent risk of death or serious physical injury, disclosure must reach an appropriate law-enforcement or public-safety authority within 24 hours. Amended reports may be filed at any time, and developers may voluntarily report incidents involving non-frontier foundation models. Reports are exempt from the public-records law. The Attorney General publishes an annual anonymized and aggregated report to the clerks of both chambers and the governor, and may transmit individual reports to the legislature, the governor, or federal or state agencies.
A federal-equivalence safe harbour lets the Attorney General designate by regulation federal laws, regulations or guidance whose incident-reporting standards are "substantially equivalent to, or stricter than" the chapter's. A developer declaring intent to comply through a designated federal instrument is deemed compliant to that extent, but failure to meet the federal standard is itself a violation of the chapter.
Third-party audits and independent evaluations
Section 3A carries the two requirements added on the Senate floor, and they operate on different clocks.
Audit (§ 3A(a)) — a large frontier developer must annually retain a third party to perform an independent audit of compliance with section 2, excluding the risk-report subsection. The auditor gets access to all reasonably necessary materials including unredacted published documents, subject to reasonable security protocols such as restrictions on note-taking, on-premises review and confidentiality terms. The report must state whether the developer substantially complied, describe material deviations with rationale and improvement recommendations, assess internal controls including the designation and empowerment of responsible senior personnel, list audit personnel, disclose conflict-of-interest procedures and any conflicts, describe methodology, and carry the lead auditor's signature. Within 30 days of receipt the developer publishes a high-level summary and a redacted copy and sends the redacted report to the Attorney General, retaining the unredacted copy for the deployment period plus five years.
Independent evaluation (§ 3A(b)) — a large frontier developer must engage at least one third party to independently evaluate its frontier models "with respect to each category of catastrophic risk." The engagement must occur "not more than 30 days after publishing each risk report," and an independent evaluation must be conducted "not less than once every 120 days." The evaluation must independently assess each type of catastrophic risk against capabilities, threat models and safeguards; assess the models' ability to automate AI research and development; and review the most recent risk report for the adequacy and completeness of its disclosures, the analytical rigour of the risk methodology, the appropriateness and materiality of redactions, and whether the evaluator disagrees with any of its claims. The evaluator gets access to unredacted materials and to "the large frontier developer's most capable frontier models," may ask relevant questions and must receive reasonable responses. A public version is published within 30 days of delivery, with the developer linking to it. Multiple third parties may divide the work by risk category or duty provided the duties are collectively performed and each states its scope.
Independence rules bar any financial, operational or management dependence on the developer or its affiliates and require freedom from control in reaching conclusions. Compensation is at reasonable market rates and may not be conditioned on results. Before accepting an engagement the third party must certify independence in writing to both the developer and the Attorney General, disclosing funding sources, other recent engagements with the developer or its affiliates, and any facts bearing on independence.
The Attorney General must implement an "independent evaluation ecosystem plan" in consultation with academic institutions, nonprofits and industry: publishing qualification standards for evaluators, exploring a licensing system, arranging or providing pooled or government funding subject to appropriation, examining the feasibility of licensing evaluators for sub-catastrophic harms and reporting to the joint committees on advanced information technology and on economic development, and funding nascent evaluator organizations. The Attorney General may also publish a rating system for qualified third parties.
Enforcement and penalties
A large frontier developer that fails to publish or transmit a compliant document, makes a statement in violation of the chapter, fails to report an incident, or fails to comply with its own frontier AI framework faces a civil penalty of not more than $1,000,000 for a first violation and not more than $3,000,000 for subsequent violations. Penalties are "recover[able] only in a civil action brought by the attorney general" — there is no private right of action for the penalty. The Attorney General may promulgate implementing regulations.
Whistleblower protections
Section 7 bars any rule, policy or contract preventing a covered employee from disclosing, and bars retaliation for disclosing, information to the Attorney General, a federal authority, a person with authority over the employee, or another covered employee with authority to investigate, where the employee reasonably believes the information shows either a specific and substantial danger to public health or safety from a catastrophic risk, or a violation of the chapter. Developers must give covered employees clear notice of these rights, by permanent workplace posting with equivalent notice to new and remote employees within 90 days, or by annual acknowledged written notice.
A large frontier developer must provide a reasonable internal disclosure process, anonymous or named, with monthly status updates to the discloser, and disclosures and responses must be shared with officers and directors at least quarterly, excluding any officer or director alleged to have engaged in wrongdoing.
The burden-shifting is unusual: once a plaintiff shows by a preponderance that a proscribed activity was a contributing factor in the adverse action, the developer must show by clear and convincing evidence that the action would have occurred for legitimate independent reasons. Courts may award attorney's fees and costs, temporary injunctive relief issues "on a showing that reasonable cause exists to believe a violation has occurred," the chilling effect on other employees is an express factor, and injunctive relief "shall not be stayed pending appeal."
Chapter 93N — chatbot disclosure
A separate, much shorter chapter requires any commercial entity deploying a chatbot to "clearly and conspicuously disclose to the person with whom the chatbot interacts that the person is interacting with a chatbot and not a human." A "chatbot" is an automated program designed to simulate conversation with human users, whether through generative AI or similar technology, by audio, visual or textual means. Violation is deemed an unfair or deceptive act under chapter 93A, which carries that statute's existing remedies.
Effective dates
| Provision | Date |
|---|---|
| Chapter 93M generally | July 1, 2027 (SECTION 355) |
| First risk report | 180 days after the act's effective date, or 180 days after first qualifying as a large frontier developer, whichever is later (SECTION 345) |
| Annual audit requirement | January 1, 2027, or 180 days after first qualifying as a large frontier developer, whichever is later (SECTION 346) |
| Independent evaluation requirement | January 1, 2027, or 180 days after first qualifying, whichever is later (SECTION 347) |
| Attorney General's evaluation-ecosystem plan | Within 1 year of the act's effective date (SECTION 348) |
| First Attorney General definitions report | By January 1, 2028 (SECTION 342) |
| First incident and covered-employee reports | By January 1, 2029 (SECTION 343) |
Sections 346 and 347 set a January 1, 2027 floor for the audit and evaluation requirements while section 355 places chapter 93M itself in effect on July 1, 2027. The engrossed text does not reconcile the two dates.
Provenance
Captured September 4, 2026 from the Massachusetts General Court's bill-text service for S.3228. Chapters 93M and 93N are reproduced in full in the raw file, together with the seven later sections setting phase-in dates for chapter 93M; the remainder of the roughly 600,000-character bond bill is not reproduced.
Relationships
- supports: Massachusetts economic development bill — AI safety provisions (2026) — primary text for that page
- related: California SB 53 — comparable 10^26 FLOP-plus-revenue coverage test
- related: New York RAISE Act, Illinois SB 315 (frontier safety framework with mandatory third-party audits) — the other state frontier-AI frameworks
- related: AI Safety Cases and Frameworks — the frontier AI framework obligation
- related: Compute Thresholds — the 10^26 operations threshold
- related: State-Level AI Regulation, AI Federalism
- related: AI Benchmarks and Evaluation — the audit and independent-evaluation regime