Bipartisan House Bill Would Set Federal Rules for Advanced AI

A new bipartisan bill called the FRONTIER Act would impose safety frameworks, outside audits and incident reporting on major AI developers, with heavy fines and possible prison time, though critics point to gaps in verification, funding and state-law preemption.

Six House members, three from each party, introduced the FRONTIER Act, drawn from an earlier draft bill written by the same lead sponsors.
Companies would write their own safety standards, but licensed outside reviewers would judge whether those standards are adequate and check that they are followed, with serious safety incidents reported within days.
A new Under Secretary for AI Security at the Commerce Department would oversee the rules, and the Commerce Secretary could order a halt to work on a model posing an imminent catastrophic risk.
Penalties reach $1 million per day, or $10 million per day for defying an emergency order, and deliberate defiance could bring prison time, which the author describes as unprecedented for an AI safety law.
Concerns raised include no deadline for licensing verifiers, dropped funding for the federal evaluation agency, no whistleblower protections, and language that would block state laws and other federal restrictions.