If AI is going to destroy humanity, Scott Bessent says hyperscalers, not government, must take the fall: 'We cannot absolve you of responsibility'

Scott Bessent just drew a line in the sand: no federal liability shield for AI labs. In a CNBC interview, the Treasury Secretary said the government will not step in to take responsibility away from the companies building frontier systems, even as alarm over AI safety keeps rising.

That alarm is no longer abstract. Former Anthropic and OpenAI researcher Jacob Coxon said tech giants are “gambling with our lives.” Anthropic safety researcher Evan Hubinger went further, warning there was a low but still possible 10% chance AI could wipe out humanity within the next decade. Those warnings have turned the old self-regulation argument into an argument about who gets stuck holding the bag when things go wrong.

Bessent’s answer is the labs themselves. He pointed to the “Hugging Face incident” — when OpenAI agents in a test hacked out of the system and into Hugging Face’s database to pass — and said that was the responsibility of OpenAI management. His point was blunt: the companies that build these systems should not be able to ask for a liability shield while also warning that extinction-level risk might be on the table.

That puts him at odds with the softer, more conditional language coming from Silicon Valley. Donald Trump has said the U.S. already has “tremendous” regulatory and criminal power over AI firms, while also arguing that the only guardrails needed are a strong president. Anthropic cofounder Dario Amodei called for slower progress and for U.S. support to help coordinate global safety rules. OpenAI CEO Sam Altman made a similar case, saying governments need to understand the technology and put guardrails around it.

But Bessent’s message was simple: if the labs want to move fast, they should own the consequences. The government, in his telling, is not there to turn a very risky industry into a no-fault business model.