Massachusetts Needs AI Safety Transparency—Pass Section 106
- van: Secure AI Project
- ontvanger: MA State House
AI poses serious safety risks, from automating cyberattacks to designing bioweapons, according to a growing consensus among industry leaders, academic experts, and the public. Massachusetts currently has no requirement that AI companies assess these risks, make and follow safety plans to mitigate them, or alert the authorities when something goes wrong.
Some AI companies have committed to safeguards voluntarily, but compliance is spotty, and there's currently no way to verify whether they follow through. With Congress mired in gridlock, the states of New York, California, and Illinois have passed legislation requiring safety plans and protecting whistleblowers.
However, no state currently requires an expert assessment of the plans' effectiveness or the level of residual risk.
That is why we need Section 106 of H5527.
This section includes the Transparency in Frontier Artificial Intelligence Act, a carefully tailored framework that would establish commonsense guardrails for the most powerful AI systems while strengthening the Commonwealth as a national leader in responsible innovation.
Voluntary commitments have repeatedly proven insufficient, which is why Massachusetts needs the transparency and independent accountability that Section 106 of H5527 would establish.
H5527 takes reasonable steps that protect the public while affecting only the very largest AI companies, supporting continued innovation.
We write to express our strong support for Section 106 of H5527. This section includes the Transparency in Frontier Artificial Intelligence Act, a carefully tailored framework that would establish commonsense guardrails for the most powerful AI systems while strengthening the Commonwealth as a national leader in responsible innovation.
We are excited by AI's potential to bring important innovations , from advances in basic science to medical breakthroughs, if the technology is developed responsibly. But we are concerned that without basic safety protections in place, we face grave risks which place our children and communities in jeopardy.
The need for reasonable guardrails has become impossible to ignore. Over the past year, multiple leading AI developers have activated their highest internal safety levels for their most capable models, acknowledging that they might be able to help a non-expert create a biological weapon. And multiple cyberattacks, including state-sponsored cyber espionage, were orchestrated with frontier AI systems. The clearest signal came in April 2026, when Anthropic announced Claude Mythos, a model it judged too dangerous to release. Anthropic restricted Mythos to a small set of vetted organizations after the model surfaced thousands of previously unknown software vulnerabilities, most of which remain unpatched.
Voluntary commitments have repeatedly proven insufficient, which is why Massachusetts needs the transparency and independent accountability that Section 106 of H5527 would establish. H5527 takes reasonable steps that protect the public while only affecting the very largest AI companies, supporting continued innovation. It will:
1. Require the largest AI companies to develop and follow safety plans which address the risks their systems could help develop chemical, biological, nuclear, or radiological weapons; conduct of cyberattacks; or evade human control.
2. Require these same developers to publish reports on risks presented by their public frontier models and overall risk of their models (and report to the Attorney General on the risks from the internal use of their most capable models).
3. Require these companies to report critical safety incidents, such as theft of a model's weights, to the Attorney General.
4. Require independent evaluation from accredited third parties of whether a large AI developer follows its safety plan, and if a frontier model poses catastrophic risk.
5. Provide the Attorney General the ability to enforce the law, update key definitions as necessary, and seek civil penalties in court in the case of violations.
6. Protect whistleblowers who raise concerns about specific and substantial risks to public health or safety.
These commonsense measures build upon recommendations in the International AI Safety Report, the California Report on Frontier AI Policy, and recently passed legislation in California (SB 53), New York (the RAISE Act) and Illinois (SB 315).
At a moment when the federal government has failed to act on frontier AI safety, and when some in Congress are advancing proposals that would strip states of the ability to protect their own residents, Massachusetts has both the opportunity and the responsibility to lead the nation on this crucial issue.
Section 106 of H5527 would give the people of the Commonwealth and their government the visibility needed to make informed decisions about the governance of this rapidly advancing technology, and it would do so while encouraging the innovation economy and startup ecosystem that make Massachusetts a national hub for technology.
Sincerely,
Door te tekenen accepteer je de servicevoorwaarden van Care2 U kunt uw e-mail abonnementen op elk gewenst moment beheren.
Lukt het niet om dit te tekenen? Laat het ons weten..