US government releases Anthropic's blocked Mythos 5 AI model to roughly 100 selected organizations - Fable 5 remains restricted
What it really says
The US government on June 27, 2026, partially lifted its export block on Anthropic's AI model Mythos 5. Commerce Secretary Howard Lutnick wrote in a letter to Anthropic that 'appropriate safeguards are in place to permit certain trusted partners to access the Claude Mythos 5 Model.' Approximately 100 US institutions, including companies and federal agencies, will receive access. Organizations include Cisco and JPMorgan Chase, which participate in the invitation-only program 'Project Glasswing,' an initiative where Mythos 5 is deployed specifically for cybersecurity purposes. Mythos 5 is Anthropic's most powerful model, possessing exceptional capabilities in cybersecurity that can find and analyze vulnerabilities in IT systems extraordinarily quickly. It was precisely these capabilities that led to an unprecedented export block by the US government in mid-June 2026, during which Anthropic was ordered to disable both Mythos 5 and the weaker model Fable 5 for all foreign nationals, including its own employees. The current partial release explicitly does not include Fable 5, which was briefly the most powerful publicly available AI model. According to people familiar with the talks, work is underway on restoring Fable access as well, but the timeline is unclear. In parallel, a lawsuit by the US Department of Defense against Anthropic continues on a separate legal track.
Our assessment
This development merits a yellow rating because it has both reassuring and concerning aspects. The reassuring side: the fact that an AI model is so capable that the US government controls its access demonstrates that AI safety is being taken seriously. The partial release only to vetted organizations for cybersecurity purposes is a responsible approach. The concerning side: first, the existence of Mythos 5 confirms that AI models now have capabilities that governments classify as national security risks. Second, the situation reveals a new form of power concentration: a few US companies and agencies gain access to the most powerful AI tools while the rest of the world, including European allies, remains locked out. Third, Fable 5, which was publicly available until recently, remains blocked, affecting millions of users worldwide. The fundamental question remains: who decides on global access to AI capabilities, and by what criteria?
Relevance for Germany
This development has immediate consequences for Germany and Europe. First, European users and companies remain cut off from Fable 5, which until mid-June was one of the most widely used AI models. German companies that had built their workflows on Claude had to switch to alternatives within days. Second, the incident demonstrates Europe's fundamental dependence on US AI infrastructure: a decision in Washington can cut off access to key technologies overnight. Third, the selective release only to US organizations raises questions of digital sovereignty that play a central role in the European debate about the EU AI Act and building indigenous AI capacities. Fourth, the cybersecurity aspect is directly relevant: if only US companies have access to the most advanced AI tools for cybersecurity, this could impair the defensive capabilities of European companies and agencies against cyberattacks, especially given the Five Eyes intelligence agencies' warning that AI-powered cyberattacks are 'months, not years' away.
Fact check
The partial release of Mythos 5 on June 27, 2026, is consistently reported by CNN, CNBC, Fortune, Axios, and Semafor. The quote from Commerce Secretary Lutnick about 'appropriate safeguards' is identically reproduced across multiple reports and originates from the official letter to Anthropic. The figure of approximately 100 institutions is confirmed by CNN and Fortune. Cisco and JPMorgan Chase's participation in Project Glasswing is reported by Semafor. The continued restriction on Fable 5 and ongoing talks about its restoration are confirmed by Axios and Fortune. The parallel DOD lawsuit is mentioned by multiple sources. The original export block from mid-June 2026 is documented through Forbes reporting from June 16.
Source
- • https://fortune.com/2026/06/27/anthropic-mythos-5-ai-model-us-commerce-department-clearance-fable/
- • https://www.cnbc.com/2026/06/26/us-government-anthropic-claude-mythos5-ai.html
- • https://www.axios.com/2026/06/27/anthropic-fable-5-return-soon
- • https://www.semafor.com/article/06/27/2026/us-releases-powerful-anthropic-model-mythos-to-some-us-companies