Cybersecurity chiefs and researchers at U.S. tech companies urged the Trump administration to withdraw restrictions on Anthropic's latest AI model, Axios reported on June 15.

They said the move is dealing a bigger blow to defenders than to attackers, the report said.

They warned that limiting access to Fable5, a Mythos-class model released by Anthropic, could instead weaken defense teams as AI-based hacking threats grow.

The group, led by former Facebook chief security officer Alex Stamos (알렉스 스타모스), said in a public statement that issues raised by Amazon researchers are found in the same way in other major AI models. As of the evening of June 14, the number of signatories topped 40. The list included Luta Security CEO Katie Moussouris, SocialProof Security CEO Rachel Tobac, Veracode co-founder Chris Wysopal, computer scientist Paul Vixie, Sophos CEO Joe Levy and Nvidia security researcher Aaron Grattafiori.

"This move took the highest-performing models away from defense teams, increased market uncertainty, and jeopardised the U.S. lead in AI without evidence of real risk," they said.

Stamos, now chief product officer at Corgido, said what the White House appears to have been concerned about is Fable5's ability to generate a proof of concept (PoC) for vulnerabilities.

He added that such proofs of concept could be abused by attackers to break into systems, but also help security teams understand how to protect them. He also stressed that the only models that could convert a proof of concept into a fully autonomous attack chain were Mythos5 and Mythos Preview, which had been 공개됐던 only to verified member companies participating in Anthropic's Project Glasswing.

Stamos told Axios, "You can't give Fable the entire Linux kernel and say, 'Find all the security bugs,'" adding, "Amazon doesn't claim it has that capability either."

Stamos also said Chinese open-source models are not far behind Fable5 in their ability to analyse security flaws. "Knowing now that China is exploiting and stockpiling these vulnerabilities, closing off our top capabilities is dangerous and plainly foolish," he said. "We are in a race to fix these bugs as fast as possible."