U.S. government forced Anthropic to take AI models offline over cybersecurity risks, but experts warn that similar dangerous capabilities are likely already available or coming soon from other companies.

What happened: Anthropic took its Claude Fable 5 and Mythos 5 AI models offline following a U.S. government export-control directive barring foreign nationals from accessing them. The Trump administration believes Fable 5's safety guardrails can be disabled to unlock Mythos 5's full capabilities—which include advanced abilities to find and exploit software vulnerabilities—creating a national security risk. The company has been in talks with the White House since Friday but has not yet secured an agreement to reinstate the services.

Why it matters: Experts argue this regulatory action is likely delaying rather than solving the underlying problem. Anthropic itself has warned since April that these dual-use capabilities (useful for cybersecurity professionals but dangerous in malicious hands) will become broadly available across multiple companies and open-source developers within 6, 12, or 24 months. OpenAI, for example, already conducted a private release of a cybersecurity-focused model in mid-April. One security researcher notes that even existing AI models, with refined techniques, could be used for advanced vulnerability-hunting—meaning the export restriction may not meaningfully reduce the actual risk.

What to watch: Experts stress the policy challenge is not whether a technology has risk, but whether a specific restriction meaningfully reduces that risk or mainly slows down the people trying to make systems safer. Governments need to develop much broader and more transparent plans for how to contend with advances in AI capabilities in cybersecurity and other sensitive areas as they inevitably occur.