POLITICO reports Anthropic has quietly pitched lawmakers a draft frontier AI safety bill that would let a court block dangerous models
BREAKING: Anthropic has quietly pitched Capitol Hill lawmakers on a draft bill to regulate frontier AI safety and security, per POLITICO The bill would let the government block a model’s release or use only if it poses a foreseeable, imminent risk of more than $1 billion in damages, or of killing or seriously injuring more than 50 people.
Context
An AI lab writing its own federal rules could shape how the most powerful models are tested and when the government can stop one, and critics say it would also reach far smaller companies.
This is secondhand: unusual_whales is relaying a POLITICO report that cites three people with knowledge of the effort. An Anthropic spokesperson told Reuters the bill reflects proposals in the company's Advanced AI Framework, and no Anthropic post about it on October 18 was found.
Deep dive
On October 18, unusual_whales posted that Anthropic has quietly pitched Capitol Hill lawmakers on a draft bill to regulate frontier AI safety and security, citing POLITICO. It wrote that the bill would let the government block a model's release or use only if it poses a foreseeable, imminent risk of more than $1 billion in damages, or of killing or seriously injuring more than 50 people. Watcher.Guru posted a shorter version without naming a source.
POLITICO first reported the draft on October 16. According to Reuters' summary, the risk must stem from a biological weapon, a model evading its developer's controls, or a cyberattack. Independent evaluators registered with the Commerce Department would periodically assess labs and their models, could be paid by the labs if payment is not tied to the outcome, and would have to report an imminent risk to Commerce within 72 hours. Commerce could ask a federal judge in Washington, D.C. to halt a model, and no other agency could suspend one.
An Anthropic spokesperson told Reuters the bill would give the government the ability to block or deter dangerous deployments of the most powerful models, and requires developers to test frontier models, be transparent about findings and get them independently evaluated.
On October 17, David Sacks wrote that the bill defines the frontier as any model trained on 10 to the 25th FLOPs, counting fine tuning and reinforcement learning. He argued that this 2023 threshold would sweep in most of Silicon Valley and favor incumbents, and asked Anthropic to clarify whether it believes that represents the frontier.
What to expect next
Watch whether any member of Congress agrees to sponsor the draft, and whether Anthropic publishes the text or responds to Sacks on the compute threshold.
Also watch how other labs, startups and open model developers react to the 10 to the 25th FLOPs definition, and whether the Commerce Department comments on the role the bill would give it. The draft may also feed into the broader debate over how federal and state AI rules fit together.