Dozens of prominent cybersecurity experts are criticizing the Trump administration for banning Anthropic from letting foreign entities access its powerful new AI models.
The U.S. government’s export-control directive — which prompted Anthropic to entirely suspend access to its Fable 5 and Mythos 5 models, including in the U.S. — “has taken the best models away from defenders, created market uncertainty, and risked America’s AI leadership without any real risk to justify it,” the experts wrote on Sunday in a letter to Commerce Secretary Howard Lutnick and National Cyber Director Sean Cairncross.
The Trump administration on Friday told Anthropic to immediately prevent foreign nationals, including Anthropic employees, from accessing the new models, reportedly after Amazon researchers were able to circumvent some of Fable 5’s anti-hacking guardrails. Anthropic designed Fable 5 to offer the same power as its now-famous Mythos model but with restrictions meant to prevent its weaponization, but Amazon CEO Andy Jassy reportedly warned Trump administration officials that the guardrails were flawed.
Clash over seriousness of vulnerability
In announcing its decision to completely shut down access to Fable 5 and Mythos 5, Anthropic said the U.S. government informed it of “a potential narrow, non-universal jailbreak, which essentially consists of asking the model to read a specific codebase and fix any software flaws.”
“We reviewed a demonstration of this specific technique being used to identify a small number of previously known, minor vulnerabilities,” Anthropic said in its statement. “These vulnerabilities all appear relatively simple, and we have found that other publicly-available models are able to discover them as well without requiring a bypass.”
The cybersecurity experts’ letter — which as of Monday morning carried 76 signatures, including CEOs, CISOs, venture capitalists and prominent security researchers — echoed Anthropic’s point.
“Anthropic’s Mythos-class models are quite good at finding flaws and weaponizing exploits,” they wrote. “However, they are not uniquely good at these tasks, and many of the undersigned individuals regularly use other foundation and open-source models for security audits and red-teaming every day.”
Indeed, a recent government competition spurred the creation of open-source AI models that have revolutionized the process of finding software vulnerabilities, in many cases with significantly lower costs than leading AI vendors’ better-known models.
U.S. AI competitiveness concerns
Anthropic argued that it made little sense to ban access to Fable and Mythos based on a capability that many models possessed.
“If this standard was applied across the industry,” Anthropic said, “we believe it would essentially halt all new model deployments for all frontier model providers.”
The battle over access to Anthropic’s AI models comes as the U.S. and China tussle for supremacy in an industry seen as increasingly vital to everything from machine-speed cyber defense to cutting-edge medical research.
China’s advanced AI models are “only months behind the best American models,” the cybersecurity experts said in their letter. “To pull the best capabilities away from defenders without a good reason when our adversaries are rapidly advancing is dangerous.”
