Anthropic's Fable 5 restored worldwide after 19-day US export ban - new safety classifier aims to prevent misuse
What it really says
On July 1, 2026, Anthropic restored worldwide access to its AI model Claude Fable 5 after the US Department of Commerce lifted an export control order the previous evening that had been in effect since June 12. The 19-day ban was triggered by Amazon researchers who discovered a jailbreak technique that allowed Fable 5 to bypass its safety measures and identify software vulnerabilities as well as generate exploit code. Amazon CEO Andy Jassy reported the finding to US authorities. For the restoration, Anthropic implemented several safety measures: a new classifier that blocks the reported technique in over 99 percent of cases, with flagged requests automatically rerouted to the model Claude Opus 4.8, which has stricter safety boundaries. Anthropic also accepted undisclosed conditions from the US government and established a HackerOne channel for responsible disclosure of security vulnerabilities. At the same time, Anthropic acknowledges that the new classifier also more frequently blocks benign requests during programming and debugging tasks. Reduced usage limits apply until July 7. Simultaneously, Claude Sonnet 5, a new mid-tier model, was introduced as the default for all free and Pro users worldwide. The most powerful model, Mythos 5, remains available only to approximately 100 vetted US organizations through the 'Project Glasswing' program.
Our assessment
This development merits a yellow rating because it has both reassuring and concerning aspects. The reassuring side: the 19-day ban demonstrates that the US government actually intervenes on AI safety risks, even when this affects a major US company and millions of users worldwide. The new safety measures, particularly the classifier with over 99 percent blocking rate and automatic rerouting to a more secure model, represent tangible progress. Anthropic's establishment of a public bug bounty channel strengthens independent security research. The concerning side: the ban revealed how vulnerable companies and users are when they depend on a single AI infrastructure. Within hours, millions of people worldwide lost access to a tool they had integrated into their workflows. The episode also raises questions: a single jailbreak finding by a competitor (Amazon) was enough to trigger a worldwide shutdown. The fact that Anthropic had to accept undisclosed government conditions reveals a new dimension of state control over AI companies. And Mythos 5 remaining available only to US organizations solidifies a two-tier structure in access to the most powerful AI models.
Relevance for Germany
For Germany and Europe, the restoration of Fable 5 brings immediate relief but also serves as a warning. German companies, developers, and researchers who use Claude as a work tool were cut off from a central AI model for 19 days. Many had to hastily switch to alternatives like GPT-5.6 or Gemini, incurring costs and productivity losses. The episode underscores the call for digital sovereignty that is growing louder in German and European debate: as long as Europe does not operate its own AI models at this performance level, it remains dependent on decisions made in Washington. The fact that the ban was not triggered by a security problem in Europe, yet Europe had to bear the consequences, makes this dependency particularly stark. The new safety classifier directly affects German users as well: those using Fable 5 for programming or technical tasks must expect more frequent false alarms. On the positive side, the improved protection against misuse of cyber capabilities also benefits German companies and government agencies that are potential targets of AI-powered cyberattacks.
Fact check
The restoration of Fable 5 on July 1, 2026, is confirmed by Anthropic's official announcement and consistent reporting from Forbes, The Hacker News, CoinDesk, Digital Trends, and MarketScale. The 19-day duration of the ban (June 12 to June 30) is documented by Forbes and MarketScale. The trigger by Amazon researchers and the report by CEO Andy Jassy are reported by The Hacker News and CoinDesk. Technical details of the new classifier - over 99 percent blocking rate, rerouting to Opus 4.8, more frequent false positives for benign requests - originate from Anthropic's official announcement and are confirmed by TechTimes. The reduced usage limits until July 7 are mentioned by Digital Trends. The continued restriction of Mythos 5 to US organizations is evidenced by reporting from June 27 (Fortune, CNBC) and the current Anthropic announcement.
Source
- • https://www.anthropic.com/news/redeploying-fable-5
- • https://www.forbes.com/sites/ronschmelzer/2026/06/29/is-anthropics-fable-5-coming-back-this-week/
- • https://thehackernews.com/2026/07/anthropic-restores-claude-fable-5-after.html
- • https://www.coindesk.com/tech/2026/07/01/anthropic-restores-ai-models-fable-mythos-after-the-u-s-lifts-export-controls