US Approves Anthropic's Mythos 5 Cybersecurity Redeployment

Anthropic has received approval from the U.S. government to bring its most powerful cybersecurity AI model, Claude Mythos 5, back online for a select group of American organizations responsible for protecting critical infrastructure. The decision, confirmed by Commerce Secretary Howard Lutnick on Friday, June 26, 2026, partially reverses a federal order issued on June 12 that had forced Anthropic to suspend the model entirely.

Roughly 100 organizations, including government agencies and private companies in sectors such as energy, banking, and telecommunications, are expected to regain access in the coming days. Wider availability, along with the return of the companion model Claude Fable 5, remains under discussion between Anthropic and federal officials.

What Triggered the Original Shutdown

The restrictions trace back to an export control directive the Commerce Department sent Anthropic on June 12 at 5:21 p.m. ET. The order barred foreign nationals from accessing sensitive AI technology — a category that, in practice, covered both Mythos 5 and Fable 5.

Because foreign-national employees worked at Anthropic itself, as well as at several partner organizations that had been testing the models, the company determined it had no way to comply with the order short of pulling the systems offline for everyone, domestic and international users alike.

A Model Anthropic Initially Held Back

Mythos 5 was not a routine product launch. Anthropic had already treated the model with unusual caution months earlier, withholding a preview version from public release in April and instead sharing it only with a small group of more than 50 vetted partners through an effort the company calls Project Glasswing.

That caution stemmed from the model's capabilities. According to Anthropic, the system could autonomously hunt for software vulnerabilities at a scale and speed previous tools couldn't match — reportedly surfacing thousands of high-severity flaws across major operating systems and browsers, including a vulnerability in OpenBSD that had gone undetected for 27 years and a 16-year-old flaw in FFmpeg.

How the Restoration Will Work

Lutnick informed Anthropic in a letter that the government was satisfied with the safeguards the company had put in place for trusted users, according to a person familiar with the letter's contents. The restored access applies specifically to organizations that operate or defend U.S. critical infrastructure — not the general public.

Anthropic confirmed the development in a post on its official X account, stating it was working to restore access for approved organizations "quickly" and that conversations with the government would continue over the weekend.

• Who qualifies: Roughly 100 vetted U.S. organizations, spanning government agencies and private firms

• What's restored: Claude Mythos 5 only, for defensive cybersecurity use

• What's still pending: General public access to Mythos 5, and any access to Claude Fable 5

Fable 5 Remains in Limbo

Claude Fable 5, which Anthropic has described as its most capable general-use model, was swept into the same June 12 shutdown but has not been part of the restoration announced this week. A person close to the company said Anthropic intends to keep pressing federal officials over the weekend for a path to bring Fable 5 back, though no timeline has been set.

Industry Reaction Has Been Mixed

The episode has drawn criticism from multiple directions. John Coleman of the Foundation for Individual Rights and Expression argued the government's review process lacked transparency and gave federal officials outsized influence over which companies get to use frontier AI tools. OpenAI chief executive Sam Altman separately objected to the idea of government agencies effectively selecting which customers a private AI company can serve.

The timing has also drawn attention because of a parallel development at a rival firm. Hours before Lutnick's letter to Anthropic became public, OpenAI announced it would roll out its newest model family in stages at the federal government's request, rather than the broad release it had originally planned.

Part of a Broader Federal Push on AI Oversight

The Mythos 5 dispute unfolded against the backdrop of a presidential executive order issued earlier in June directing federal agencies to strengthen cyber defenses and create a formal mechanism for safety-testing advanced AI models before they reach the public. Anthropic had sent senior engineers and scientists to Washington to work directly with the Commerce Department and the Office of the National Cyber Director in the weeks following the shutdown.

What Comes Next

Anthropic has not disclosed exactly when the roughly 100 cleared organizations will see access restored, only that the rollout is already underway. The company said it will continue pressing for two further outcomes: a broader expansion of Mythos 5 to additional critical infrastructure operators, and a separate resolution that would let Fable 5 return to general use.

For now, the bulk of Anthropic's customer base — including international partners that had been part of the original Project Glasswing cohort in countries such as Canada, Germany, Japan, and South Korea — remain without access to either model while talks continue.

[Source: Bloomberg]