The Trump administration has lifted export controls on Anthropic's powerful AI models, Claude Fable 5 and Mythos 5, allowing public access to resume after an 18-day restriction. This decision, announced by Commerce Secretary Howard Lutnick, follows close collaboration between Anthropic and the government to implement new safeguards addressing security concerns, particularly around potential "jailbreaks" that could allow misuse of the models.
Anthropic had abruptly shut off access to Fable 5 and Mythos 5 last month after President Donald Trump's administration ordered the company to restrict foreign nationals, including employees, from accessing the models. The primary concern stemmed from reports that researchers were able to "jailbreak" Fable 5, potentially enabling it to automate sophisticated cyberattacks or accelerate biological weapons development. Anthropic stated that its new safeguard blocks these jailbreaks 99% of the time, and in the remaining cases, the output only involved previously discovered or already patched security flaws.
As part of the agreement, Anthropic will provide pre-release government access and evaluation for future models relevant to national security and engage in rapid information sharing regarding significant jailbreaks or misuse patterns. The company also plans to collaborate with Amazon, Microsoft, and Google on a framework for assessing AI jailbreaks. While the new safeguards might lead to more benign requests being flagged, Anthropic is committed to refining these classifiers to reduce false positives.
Analysts noted that this move sets a precedent for the ad hoc regulation of frontier AI models, with the US government seeking to influence their release. Other AI companies, like OpenAI, have also seen their models, such as GPT-5.6, rolled out to only a select group of approved customers following government requests. The Trump administration faces an August deadline to create standardized benchmarks for evaluating the security risks of new AI models, highlighting the ongoing evolution of AI regulation.
Anthropic expressed gratitude to its users for their patience and to all who worked on redeploying the models, with White House chief of staff Susie Wiles commending the unprecedented cooperation between the government and the private sector. Starting Wednesday, access to Fable 5 and Mythos 5 will begin to be restored globally on Claude platforms, with subsequent re-enablement on AWS, Google Cloud, and Microsoft Foundry.
