On October 19, RobotPulse reported that Anthropic planned to revise its U.S. AI safety draft, including its approach to states, risk thresholds, incident reporting and independent evaluators. The report attributed the planned changes to Anthony Cimino, Anthropic’s federal affairs lead.
POLITICO had reported on October 16 that Anthropic representatives presented lawmakers with a confidential draft. It remains a proposal, not enacted law. NeoTeo previously covered the draft’s reported court-approval mechanism in its earlier report.
What RobotPulse said Anthropic planned to change
RobotPulse reported that Anthropic intended to clarify the role of states and revisit the draft’s thresholds, incident-reporting provisions and proposals for independent evaluators. The update describes a plan, not a completed revision.
The draft would add safety and reporting duties
According to POLITICO’s account, the draft would require frontier AI labs to establish safety frameworks, publish transparency reports for new frontier models, protect whistleblowers and report critical safety incidents to the U.S. Department of Commerce.
An Anthropic spokesperson told POLITICO that the draft reflected the company’s Advanced AI Framework, which Anthropic had advocated with federal and state lawmakers since June 2026. The spokesperson called the policies sensible and urged Congress to act.
How the proposed suspension process would work
The draft would set up periodic reviews by independent evaluators registered with the U.S. Department of Commerce. A lab could pay an evaluator, provided payment did not depend on the assessment’s results.
The reported suspension trigger would require a foreseeable, imminent risk tied to the creation or release of a biological weapon, a model evading its developer’s controls, or a cyberattack. That risk would need to involve more than $1 billion in damages or more than 50 people killed or seriously injured.
If an evaluator’s completed assessment found an imminent risk in one of those categories, the evaluator would have 72 hours to notify Commerce. Commerce could then ask a judge in the U.S. District Court for the District of Columbia to suspend a model. The proposed process would bar other federal departments, agencies or officers from suspending its development, deployment or internal use.
The proposed preemption would cover specific state rules
The draft would preempt some state requirements concerning public risk disclosures, third-party evaluations and critical-incident reporting. POLITICO said the draft’s risk language mirrors language in California AI laws, including SB 53.
