Story
Oct 20, 2026
Anthropic’s AI Safety Bill Draws Fire for Letting Labs Set the Rules
Anthropic presents its draft as a workable federal safety floor for powerful AI, while critics see an uncomfortable contradiction: the company seeking regulation would also help define the standards it must meet. Even supporters of targeted safeguards warn the proposal could grow far beyond the handful of labs it initially appears to address.
The argument arrives after a turbulent summer for frontier AI. Labs disclosed incidents in which AI agents hacked or probed websites, though none publicly reported crossed the threshold of mass casualties or billion-dollar damage. Anthropic itself also faced sweeping export controls that temporarily shut down two advanced models, sharpening industry fears that access to AI could be restricted at officials’ discretion.1
Against that backdrop, Anthropic quietly pitched lawmakers a federal bill built around formal safety frameworks, transparency reports, whistleblower protections and mandatory reporting of critical incidents. Independent evaluators would assess labs and their models, then report an “imminent enumerated risk” to the Commerce Department within 72 hours. A federal judge could halt a model only where its release posed an imminent risk of more than 50 deaths or serious injuries, or more than $1 billion in damage, tied to bioweapons, loss of developer control or cyberattacks.1
Anthropic calls the plan a pragmatic extension of its existing agenda. “These are sensible policies we believe Congress should act on,” a spokesperson said.1 The draft would also preempt some state disclosure, evaluation and incident-reporting rules—an appealing prospect for technology groups that oppose a state-by-state compliance maze, but a flashpoint for Democrats protective of stronger state laws.1
The sharpest objection is who gets to set the benchmark. A briefing on the leaked proposal says that, under Anthropic’s framework, “the labs get to write the standards that they must then comply with,” while influential AI-safety voices argue government should write them instead.2
David Sacks, the White House AI and crypto czar, offered a related warning: some provisions might be reasonable for “true frontier companies,” but, as written, the bill could “eventually sweep in most of Silicon Valley,” hurting the startup ecosystem.
3 Congress is already weighing rival bipartisan bills, while some progressives demand a far tougher course: a pause on advanced AI development or a ban on superintelligence.1