Trump Administration Considers AI Model Safety Watchdog

Comparable regulatory frameworks commonly turn safety evaluation from an internal governance process into an externally reviewable one. That can affect how AI teams document model capabilities, retain evaluation evidence, define release gates, and manage access controls, even before a formal rule takes effect.

Bloomberg reports that the Trump administration is considering an independent regulator to vet the safety of artificial intelligence models. According to people familiar with the matter, the proposed agency would report to the Securities and Exchange Commission and be modeled on the Financial Industry Regulatory Authority, or FINRA. Treasury Secretary Scott Bessent helped develop the proposal, Bloomberg reports.

The proposal is not public. Bloomberg reports that White House Chief of Staff Susie Wiles is reviewing it, that President Donald Trump had not reviewed it at the time of reporting, and that the framework remains under deliberation and is subject to change.

A FINRA-like arrangement would differ from broad principles-based AI guidance because the core question would become operational evidence: what tests were run, what failure modes were found, who reviewed the results, and what conditions governed release. In similar regulated settings, reproducible evaluation records, versioned model artifacts, incident reporting, and audit trails become important engineering outputs alongside model quality.

The reporting leaves open whether any review would cover only the largest general-purpose models, specific high-risk capabilities, or downstream applications. Observers following the proposal can watch for definitions of covered models, the role of third-party evaluators, whether standards are voluntary or binding, how export-control authorities interact with the new body, and whether safety assessments require disclosure of benchmark methodology or results.