For makers and artists relying on cutting-edge tools, the landscape is shifting. The US government has ordered Anthropic to immediately disable access to Claude Fable 5 and Claude Mythos 5, citing national security risks. While Anthropic confirmed compliance via X, the company insists this move was a strategic error.
The scope of the shutdown
Received at 5:21 pm ET on Friday, the directive requires the disabling of both models globally. This extends far beyond the foreign nationals targeted by the export control order; every user worldwide is now cut off from these specific systems. Fortunately, access to Anthropic’s remaining models remains unaffected.
Why Mythos 5 matters
Mythos is Anthropic’s most powerful engine, previewed in early April and kept strictly restricted due to its uncanny ability to expose software vulnerabilities. The company reported that Mythos identified flaws in every major operating system and web browser tested. Instead of a broad release, Anthropic launched Project Glasswing, granting access to roughly 50 vetted organisations, including Amazon, Apple, Google, Microsoft, and CrowdStrike, for defensive cybersecurity work.
The commercial counterpart: Fable 5
Released just three days prior, Fable 5 was Anthropic’s response to market demand. It is a version of Mythos fitted with guardrails to block responses in high-risk areas like cybersecurity and biology, rendering it safe for general use. Benchmark tests from Vals AI confirmed it was immediately the most capable model available to the public.
The government’s justification vs. Anthropic’s rebuttal
Formally framed as an export control measure, the directive appears to stem from a claimed jailbreak of Fable 5. Anthropic argues the government has offered only verbal evidence of a “potential narrow, non-universal jailbreak” — essentially prompting the model to read a codebase and identify flaws. This capability, the company notes, is already widely available in other public models, including OpenAI’s GPT-5.5, and is routinely used by cybersecurity professionals for defensive purposes.
Anthropic maintains that its strongest safeguards operate through independent classifier systems separate from the model itself. Consequently, even if a user convinces Fable to bypass a refusal, the underlying protections against dangerous outputs remain active.
Industry reaction and irony
None of these arguments stopped the government, and Anthropic’s frustration is palpable. The company stated that recalling a commercial model deployed to hundreds of millions of people over a narrow potential jailbreak sets a dangerous precedent that could halt all new model deployments for frontier providers.
This situation creates a bitter irony for Anthropic, which has staked its public identity on being the safety-conscious alternative to rivals while pursuing an IPO this year. The very caution displayed in restricting Mythos — promoting it as a model too dangerous for public release — has seemingly attracted the exact level of scrutiny that could disrupt its business.
OpenAI’s Sam Altman is likely amused. In April, he described Anthropic’s handling of Mythos as “fear-based marketing,” noting the absurdity of selling a bomb shelter after admitting to building a bomb. While Altman did not predict a government shutdown, he identified the mechanism that has now turned against Anthropic: spending months telling the world your AI is uniquely dangerous ensures the world, including the US government, listens.
Key takeaways
- Anthropic has been forced to globally disable access to its top-tier models, Mythos 5 and Fable 5, following a US government directive citing national security concerns.
- The company argues the shutdown is disproportionate, noting that the alleged vulnerability exists in other public models like OpenAI’s GPT-5.5 and is standard practice in cybersecurity.
- Anthropic’s strategy of marketing its AI as uniquely dangerous has ironically drawn intense regulatory attention, potentially threatening its planned initial public offering.
Stay ahead of AI. Get the most important stories delivered to your inbox — no spam, no noise.
