U.S. Lifts Export Controls, Restoring Anthropic’s Claude Fable 5
Summary
Anthropic restored Claude Fable 5 worldwide after the U.S. lifted export controls. The company says a new classifier blocks a reported jailbreak technique in over 99% of attempts; Mythos 5 remains limited to selected U.S. critical-infrastructure users.
Key points
- The U.S. Commerce Department lifted export controls on Fable 5 and Mythos 5 after a review; Fable 5 returned worldwide on July 1.
- Amazon researchers reportedly found a Fable 5 jailbreak. Anthropic says it surfaced software flaws and, in one case, produced code showing how a flaw could be abused.
- Anthropic says a new classifier blocks the specific reported technique in more than 99% of attempts, but may falsely flag ordinary coding and debugging requests.
- Mythos 5 access remains restricted to roughly 100 U.S. companies and federal agencies defending critical infrastructure, according to the article.
- Anthropic is working with industry partners on a framework to assess jailbreak severity and says it will monitor reports around the clock.
- Anthropic opened a HackerOne program for reports of Fable 5 jailbreaks and agreed to provide the U.S. government earlier access to test future frontier models.
Article Details
- Event Type
- Worldwide restoration of Claude Fable 5 after the U.S. Commerce Department lifted export controls imposed following a reported jailbreak finding.
- Impact
- The controls had briefly taken Fable 5 and Mythos 5 offline for everyone because Anthropic could not reliably verify users’ nationality in real time. Anthropic says a new classifier blocks the reported jailbreak technique in more than 99% of attempts, though it causes more false alarms on ordinary coding and debugging requests. Mythos 5 remains limited to roughly 100 U.S. companies and federal agencies defending critical infrastructure. Anthropic also agreed to identify security problems, coordinate on future launches, and report malicious use to the U.S. government.
People
Andy JassyAmazon CEO; reports said Amazon's research and his concerns helped drive the original export-control order.Dario AmodeiAnthropic CEO whom reports said was not leading the negotiations; the article notes his clashes with the administration.David SacksFormer AI czar who accused Anthropic of prioritizing continued consumer-model availability over safety.Francesco BailoUniversity of Sydney AI governance researcher who told Al Jazeera that the reversal appeared to show the government conceding it had gone too far.Howard LutnickCommerce Secretary who signed off on lifting the export controls; he said the department reviewed the models with Anthropic for two weeks.Tom BrownAnthropic co-founder reportedly led negotiations with the government over the models and export controls.
Vendors
AmazonThe dispute centered on a jailbreak found by Amazon researchers, and Anthropic responded by adding a new classifier to block the technique while keeping Mythos 5 under tighter restrictions.AnthropicAnthropic has restored Claude Fable 5 worldwide after the U.S.GoogleWith Amazon, Microsoft, Google, and other partners, it wants to score each one on four things:MicrosoftWith Amazon, Microsoft, Google, and other partners, it wants to score each one on four things:OpenAIIt says the same requests work on plenty of weaker models too, including its own Claude Opus 4.8, OpenAI’s GPT-5.5, and China’s Kimi K2.7.
Products
Claude Fable 5Anthropic has restored Claude Fable 5 worldwide after the U.S.Claude Opus 4.8It says the same requests work on plenty of weaker models too, including its own Claude Opus 4.8, OpenAI’s GPT-5.5, and China’s Kimi K2.7.GPT-5.5It says the same requests work on plenty of weaker models too, including its own Claude Opus 4.8, OpenAI’s GPT-5.5, and China’s Kimi K2.7.GPT-5.6Days earlier, OpenAI previewed GPT-5.6 to a small, government-approved group rather than the public, citing the same dual-use worry: a model good enough to help defenders patch bugs is also good enough to help attackersKimi K2.7It says the same requests work on plenty of weaker models too, including its own Claude Opus 4.8, OpenAI’s GPT-5.5, and China’s Kimi K2.7.Mythos 5The dispute centered on a jailbreak found by Amazon researchers, and Anthropic responded by adding a new classifier to block the technique while keeping Mythos 5 under tighter restrictions.