US Lifts Export Controls on Fable 5 and Mythos 5 AI Models

Anthropic resumed the rollout of its Fable 5 and Mythos 5 AI models that fell under export controls imposed earlier this month over cybersecurity concerns. The distribution will start after the US Commerce Department lifted export controls.

Fable 5 launched globally from 1 July across Anthropic’s Claude platforms, with availability on AWS, Google Cloud, and Microsoft Foundry to follow. The company added it is working with the US government to broaden access to the advanced cybersecurity-focused Mythos 5 model beyond its current group of approved organisations participating in Project Glasswing.

The reversal comes less than three weeks after US President Donald Trump’s administration ordered Anthropic to suspend access to its most powerful AI models by foreign nationals, citing national security concerns related to a potential jailbreak. At the time, Anthropic called the suspension a misunderstanding.

Anthropic explained the restrictions were lifted after it introduced new protections designed to prevent users from bypassing Fable 5’s cybersecurity safeguards. The AI player worked with government agencies and partners including Amazon to review the reported issue, developing a new classifier that blocks the identified jailbreak technique in more than 99% of cases.

As a result of the update, Anthropic acknowledged the model may deliver more false positives, noting that that would be frustrating for users, but highlighted that the trade-off was necessary in the interest of making the model’s other capabilities widely available. However, the company emphasised the initial reported issues did not expose any unique Mythos-level cyber capabilities and were unlikely to be dangerous, but are nonetheless blocked by the safeguards out of an abundance of caution.

It added researchers from the US Department of Commerce’s Centre for AI Standards and Innovation tested the previous and updated safeguards and agreed they were both extraordinarily strong. The startup also outlined plans to deepen collaboration with the US government on frontier AI security, including early access to new models, faster information sharing on significant jailbreaks, and joint research on AI safety. In addition, Anthropic proposed a standardized framework for assessing the severity of AI jailbreaks, arguing that there is currently no consensus in the AI industry.