Anthropic's safety warnings may have just backfired - the government has pulled the plug on its most powerful AI

Overview

The U.S. government on Friday ordered Anthropic to immediately shut off worldwide access to two of its most powerful AI models, Claude Fable 5 and Claude Mythos 5, citing national security concerns. Anthropic said it complied but disagreed strongly, arguing the underlying concern is a claimed narrow jailbreak of Fable 5 that amounts to capability already available in other public models like OpenAI's GPT-5.5. The episode is notable because Anthropic had restricted Mythos for being exceptionally good at finding software vulnerabilities, and that very caution now appears to have drawn the scrutiny that disrupted its business.

Key Takeaways

The U.S. government ordered Anthropic on Friday to immediately shut off access to Claude Fable 5 and Claude Mythos 5 over national security concerns.
Anthropic complied but said the directive was wrong; access to its other models is not affected.
Although framed as an export control aimed at foreign nationals, the directive disabled both models for all users worldwide.
Anthropic says the real concern is a claimed narrow, non-universal jailbreak of Fable 5 involving reading a codebase to identify software flaws.
Mythos was so capable at finding security flaws that Anthropic restricted it to roughly 50 vetted organizations through Project Glasswing.
Fable 5, released three days earlier with guardrails, was the most capable public model per Vals AI benchmark tests.

Stats & Key Facts

#Anthropic said it received the directive on Friday at 5:21 pm ET.
#Mythos was shared with roughly 50 vetted organizations through Project Glasswing.
#Fable 5 was released just three days before the shutdown.
#Anthropic previewed Mythos in early April.

What the government ordered

The directive targeted Anthropic's two most powerful models.

›The U.S. government on Friday ordered Anthropic to immediately shut off access to Claude Fable 5 and Claude Mythos 5.
›The order cited national security concerns.
›Anthropic announced on X that it complied, while making clear it thinks the government got this wrong.

Anthropic said it received the directive on Friday at 5:21 pm ET. The order forced the company to disable both models for all users worldwide, not just the foreign nationals the export control order was nominally aimed at, while access to Anthropic's other models was unaffected.

Why Mythos was so tightly controlled

›Mythos is Anthropic's most capable AI model, previewed in early April.
›Anthropic kept it tightly restricted because of its exceptional ability to find security vulnerabilities in software.
›According to Anthropic, Mythos identified flaws in every major operating system and web browser it tested.

Rather than release Mythos broadly, Anthropic launched a controlled program called Project Glasswing, sharing it with roughly 50 vetted organizations, including Amazon, Apple, Google, Microsoft, and CrowdStrike, for defensive cybersecurity work.

What Fable 5 was

Fable was the commercial answer to Mythos.

›Fable 5 was released just three days before the shutdown.
›It is a version of Mythos fitted with guardrails that block responses in high-risk areas like cybersecurity and biology.
›Anthropic argued that made it safe enough for general release.

On release, Fable 5 was immediately the most capable AI model available to the public, according to benchmark tests from Vals AI, a company that tracks AI performance. It was the company's response to the obvious commercial pressure to ship a frontier model.

The disputed jailbreak

›The directive is framed as an export control action restricting foreign national access.
›Anthropic says its understanding is that the underlying concern is a claimed jailbreak of Fable 5.
›Anthropic says the government provided only verbal evidence of a potential narrow, non-universal jailbreak.

As Anthropic describes it, the jailbreak amounts to prompting the model to read a specific codebase and identify software flaws, a level of capability already widely available in other public models, including OpenAI's GPT-5.5, and used routinely by cybersecurity professionals for defensive purposes. Anthropic adds that its strongest safeguards run through independent classifier systems separate from the model, so dangerous outputs remain blocked even if someone gets the model to keep talking past a refusal.

Anthropic's pushback

›Anthropic said it disagrees that a narrow potential jailbreak should justify recalling a model deployed to hundreds of millions of people.
›It warned that applying this standard across the industry would essentially halt all new model deployments for all frontier model providers.
›The company is widely expected to pursue an IPO this year.

Anthropic has staked much of its public identity on being the safety-conscious alternative to its rivals. The article notes the irony that the caution it showed in restricting Mythos, promoted as too dangerous to release publicly, now appears to have attracted the government scrutiny that could disrupt its business most.

Rivals and reactions

›OpenAI's Sam Altman is described as likely enjoying the situation.
›In April, Altman told podcaster Ashlee Vance that Anthropic's handling of Mythos amounted to fear-based marketing.
›The disagreement highlights how safety positioning has become a competitive issue among model providers.

The reference to Altman frames the episode within the rivalry among leading AI companies. Anthropic's safety-first branding is both its differentiator and, in this case, a source of the scrutiny it faces.

Frequently Asked Questions

What did the U.S. government order Anthropic to do?

It ordered Anthropic on Friday to immediately shut off access to Claude Fable 5 and Claude Mythos 5, citing national security concerns. Anthropic complied while disagreeing.

Did the order affect users beyond foreign nationals?

Yes. Although framed as an export control aimed at foreign nationals, the directive forced Anthropic to disable both models for all users worldwide. Other Anthropic models were unaffected.

Why was Mythos restricted before this?

Anthropic restricted Mythos because of its exceptional ability to find software vulnerabilities, sharing it with roughly 50 vetted organizations through Project Glasswing for defensive cybersecurity work.

What is the claimed jailbreak?

Anthropic says the government cited a potential narrow, non-universal jailbreak of Fable 5 that amounts to prompting the model to read a codebase and identify software flaws, a capability it says is already available in other public models like GPT-5.5.

Why does Anthropic say the recall is an overreaction?

It said a narrow potential jailbreak should not justify recalling a model deployed to hundreds of millions of people, and warned that such a standard would essentially halt all new model deployments across frontier providers.

The shutdown of Fable 5 and Mythos 5 shows how Anthropic's safety-first restrictions on its most capable models drew the exact government scrutiny that has now disrupted its business.