Experts Say White House Export Ban Risks Adoption Boost for China’s AI Alternatives
Cybersecurity experts are urging White House officials to lift export controls imposed on artificial intelligence firm Anthropic’s Fable 5 and Mythos large language models.
See Also: AI Impersonation Is the New Arms Race-Is Your Workforce Ready?
More than 150 cybersecurity experts and executives, including the CEOs of Adobe, Sophos and Zoom, on Sunday signed an open-letter calling on the Trump administration to lift its export controls on dual-use technologies imposed late Friday by U.S. Commerce Secretary Howard Lutnick (see: US Pulls the Plug on Anthropic’s Top AI Models).
Signatories said the controls appeared to be triggered by research that used the models to find insecure code. But such capabilities “can be replicated on GPT-5.5, Opus, Sonnet and even Chinese models like Kimi 2.7,” they wrote.
Anthropic on Friday blocked access worldwide to Mythos 5. The San Francisco-based firm also yanked access to Fable 5, a version of Mythos it publicly released on June 9 with guardrails designed to protect against illicit use.
While Mythos has additional capabilities, “Chinese open-weight models are only months behind the best American models, and those are the models we know about,” which leaves organizations in America and allied nations at a disadvantage,” letter signatories said. “To pull the best capabilities away from defenders without a good reason when our adversaries are rapidly advancing is dangerous.”
Veteran bug hunter Katie Moussouris, founder and CEO of Luta Security, reviewed at Anthropic’s request the third-party research paper that apparently triggered the ban, and said it detailed how prompts can be used to bypass Fable 5 guardrails in pursuit of code fixes.
“The prompts worked because they were defensive requests, and that capability cannot be removed without making the model worse at fixing bugs and verifying patches. The same holds for every capable AI model, including the foreign and open-weight systems the United States cannot reach with export controls, many of which will match Fable and Mythos capabilities within months,” she said.
Anthropic and the Trump administration have continued to tussle publicly and in courtrooms, after the startup refused to give the U.S. military unrestricted access to its latest frontier models (see: DoD Says No to Anthropic Request for Reversing Blacklisting).
Anthropic representatives have reportedly been in close contact with Washington officials, seeking a solution to the export controls. Anthropic CEO Dario Amodei is attending this week’s G7 summit in Èvian-les-Bains, France, alongside U.S. President Donald Trump.
Expect the White House to soon “clarify” the “unsustainable” export restrictions, since “Anthropic can’t work like this and if they and their peer companies lose, China wins,” and “America will not stand for that,” said Ciaran Martin, the founding CEO of Britain’s National Cyber Security Centre, who’s now a professor of practice at Oxford University’s Blavatnik School of Government.
What longer-term diplomatic fallout the export ban will have remains to be seen, since the White House appears “to have tried to weaponize its AI power against its allies as well as its foes,” by failing to differentiate between the likes of Britain and Japan on the one hand, and China and Russia on the other, Martin said. “That matters.”
The incident is also being closely watched by the many businesses outside the United States that are weighing how they adopt AI models, and attempting to balance features and control against geopolitical and other availability risks.
AI Sovereignty Concerns
“The irony is that an export-control action meant to protect U.S. AI advantage may accelerate global migration away from U.S. AI,” said cybersecurity expert Mikko Hypponen, chief research officer at Finnish anti-drone company Sensofusion, in a post to LinkedIn.
“This move will push developers outside the U.S. toward Chinese models. Not because they prefer China, but because they want sovereignty. And they’re not too worried about China, because the models are open source: you can audit them as closely as you want and run them offline,” he said.
The risk that an organization’s access to AI models might be disrupted is clearly not hypothetical. “The action illustrates the risks for nations and organizations of relying on foreign sovereign AI in an increasingly fragmented geopolitical landscape,” said Nicola Cain, CEO and principal consultant at London-based Handley Gill.
At a governmental level, she said that for non-U.S. governments and organizations, the White House’s export ban “reinforces the need for measures like the EU’s tech sovereignty package and the U.K.’s Sovereign Venture Fund,” both of which are designed to promote trusted and domestic AI options, not least on national security grounds (see: US Anthropic Export Controls Sparks Sharp EU Reaction).
For individual organizations, the ability of a foreign government to restrict access to its domestically produced LLMs further demonstrates the need “for reliance on interoperable models with established off-boarding procedures,” together with an AI sovereignty program designed to balance access to AI features, functionality and cost with business resilience concerns, Cain said (see: Managing Risk and Resilience: The AI Sovereignty Imperative).
Marketing, LLMs and Mythos
While Mythos has been lauded for having an unprecedented ability to build effective exploit chains, sometimes from low-severity flaws, as the open-letter notes, many LLMs have a proven ability to spot all manner of fresh bugs in widely-used code bases, sometimes also more quickly and at less cost. “You don’t need Mythos to find vulnerabilities,” said Jaya Baloo, COO and CSO of startup Aisle, which uses AI to autonomously identify and remediate software vulnerabilities (see: Mythos Shutdown Contains a Message: Don’t Wait for Mythos).
Marketing hubris by Anthropic, which on June 1 submitted a confidential initial public offering proposal to the U.S. Securities and Exchange Commission, may now be partly to blame for the situation in which the startup now finds itself.
“If you spend a bunch of time telling people how dangerous your technology is, don’t be surprised when some of them agree with you,” said Jacob Williams, a former member of the U.S. National Security Agency’s hacking team.
