The US lifts export bans on Anthropic’s Fable 5 and Mythos 5 AI models after new safety measures. Here’s what changed and why it matters
The United States has removed export restrictions on Anthropic’s latest AI models, Fable 5 and Mythos 5. The decision comes less than three weeks after the company was ordered to suspend access over national security concerns.
Earlier in June, the US government feared the models could be misused by countries like China and Russia. As a result, Anthropic blocked access because it could not verify users’ nationalities in real time. However, the company has now added stronger safeguards. Fable 5 will return for public use from Wednesday. Meanwhile, Mythos 5 will gradually reach more trusted local and international partners through the Glasswing programme.
Anthropic teams up with Amazon, Microsoft and other partners
Anthropic is also working more closely with the US government. The company will give selected government agencies early access to future AI models. In addition, it is teaming up with Amazon, Microsoft, Google, and other partners to create shared standards for detecting and fixing AI jailbreaks. These jailbreaks are tricks that bypass an AI model’s safety rules.
US Commerce Secretary Howard Lutnick said the government could restore restrictions if Anthropic fails to keep its promises. He added that officials “reserve the right to reevaluate” the decision if circumstances change. The company revealed that Amazon researchers recently found a way to bypass Fable 5’s safeguards. The flaw allowed the model to identify software vulnerabilities and even generate exploit code in one case.
Anthropic says it has fixed that issue with a new safeguard. Any risky request will now be redirected to its safer Opus 4.8 model. Although this may block more harmless requests, the company believes the trade-off improves overall safety.
Anthropic also admitted it is “probably impossible” to make any AI model completely immune to jailbreaks. Even so, experts continue to test Fable 5 for weaknesses. Meanwhile, OpenAI is facing similar government scrutiny after delaying the wider release of GPT-5.6 at Washington’s request.
