Updated

Updated · POLITICO · Jul 27

Obernolte, Trahan Unveil AI Oversight Bill With 4 Cosponsors as House Eyes Pre-Election Action

Updated

Updated · POLITICO · Jul 27

Obernolte, Trahan Unveil AI Oversight Bill With 4 Cosponsors as House Eyes Pre-Election Action

Q: Who holds the power to define an imminent catastrophic risk that could legally force a massive AI system offline?

Under the Obernolte-Trahan bill, the practical trigger comes from independent auditors, but the legal shutdown power sits with the Commerce Department. The bill says an auditor who finds that a frontier model poses an “imminent catastrophic risk” must notify the government, after which the Commerce secretary could restrict the model’s deployment. For the most severe cases framed as a major national security threat, the bill also gives the government emergency shutdown authority. What is still unsettled is who gets to define that threshold in detail. The bill summary defines “imminent catastrophic risk” broadly as a present or impending catastrophic risk, but the more specific standards and procedures would likely be shaped through federal rulemaking by the Commerce Department’s AI security leadership created under the proposal. In other words, auditors make the technical determination in individual cases, but Commerce would control the enforceable standard and the order to limit or halt deployment. That unresolved line-drawing is one of the central policy issues. Critics of similar state laws have warned that private auditors may end up making highly subjective judgments unless the federal government sets clearer standards. So, if the bill became law, the answer would be: auditors identify the risk, but Commerce holds the authority to turn that judgment into a legally binding shutdown or deployment restriction.

2 articles · Updated · POLITICO · Jul 27

A bipartisan House bill introduced Thursday would let the Commerce secretary restrict deployment of AI models deemed an imminent catastrophic or national-security risk, while requiring large developers to publish safety reports, disclose incidents and undergo independent audits.
The revised measure narrows one of the biggest flashpoints—federal preemption of state AI laws—to three areas: catastrophic-risk transparency, independent verification and incident reporting, while leaving child safety, privacy, consumer protection and state procurement rules to states.
Trahan and Obernolte say a first documented autonomous AI breach last week underscored the need for mandatory reporting, arguing dangerous incidents now come to light only when companies choose to disclose them.
The pair want Energy and Commerce to hold a hearing or markup after the September return, but Chair Brett Guthrie has not committed to moving comprehensive AI legislation before the November midterms and GOP leaders remain wary of regulation that could curb innovation.