Anthropic announced on June 27, 2026, that the US government has authorized the company to begin redeploying Claude Mythos 5, its most advanced cybersecurity-focused AI model, to a defined group of American organizations responsible for operating and defending critical infrastructure. The decision, communicated through a letter from Commerce Secretary Howard Lutnick, marks the end of a roughly two-week suspension that had cut off access to both Mythos 5 and its safety-constrained counterpart, Claude Fable 5, for all customers under Anthropic's Project Glasswing program.

The suspension began on June 12, 2026, when an export control directive from the Commerce Department abruptly halted access to the two models. According to Anthropic, the restriction affected foreign nationals and raised concerns tied to national security, though the company did not initially disclose the precise trigger. Subsequent reporting indicated that US officials grew concerned after learning of a possible vulnerability in Anthropic's safeguards and amid broader unease about the models reaching organizations with ties to China, including reports that a South Korean telecommunications provider with Chinese links had gained access to Mythos.

In his letter, Lutnick wrote that Anthropic's work with the government since the June 12 directive had "yielded significant progress" and that the company had committed to ongoing collaboration on protocols and standards governing future releases of what the letter terms "Covered Models." The new arrangement removes the export license requirement for transferring Mythos 5 to entities named in an accompanying annex, along with their foreign national employees, as well as Anthropic's own foreign national staff working on the model. Lutnick also reserved the right to adjust or expand the approved list of partners at any time, underscoring that the government intends to maintain direct oversight of the model's distribution.

Reporting from Reuters and Semafor indicates that more than 100 companies and institutions, including a number of Fortune 500 firms and government agencies, are covered under the restored access. Anthropic said it is moving quickly to provision these approved partners and confirmed it continues to negotiate with officials to broaden the list further and to restore general availability of Fable 5, the version of the model designed with additional safeguards that route cybersecurity- and biology-related queries to other Claude models. No timeline has been given for when Fable 5 might return to wider use.

Claude Mythos first drew attention in April 2026, when Anthropic described it as a model capable of a "cybersecurity reckoning," citing its ability to autonomously discover software vulnerabilities at scale. The company has said that Mythos models, used in partnership with roughly 50 vetted organizations, have surfaced more than ten thousand high- or critical-severity flaws across widely used software, including a decades-old vulnerability in OpenBSD and a long-standing flaw in FFmpeg, along with chained Linux kernel exploits. Because of this dual-use potential, Anthropic limited distribution through Project Glasswing, a closed access program involving partners such as Amazon Web Services, Microsoft, Google, Apple, Cisco, CrowdStrike, NVIDIA, JPMorganChase, and the Linux Foundation, among others.

The episode highlights a broader shift in how frontier AI capabilities are governed in the United States. Lutnick's letter points to an emerging framework in which the Commerce Department plays a direct role in approving which organizations may access the most capable AI systems, particularly those with offensive cyber potential. The timing is notable: the restoration was announced the same day a rival lab, OpenAI, released its latest model, GPT-5.6, to a similarly limited list of government-approved partners, suggesting that case-by-case government review may become a recurring feature of how the most powerful AI models reach the market.

For now, access remains narrow. Non-US governments, international companies, and individual consumers still have no clear path to Mythos 5 or Fable 5, and some US allies have voiced frustration at being dependent on decisions made in Washington. Anthropic has said it will keep working with regulators on both expanding the approved partner list for Mythos 5 and securing a broader, general release of Fable 5, but neither outcome has a confirmed date.

Key Takeaways

The US Commerce Department has lifted part of a June 12 export control directive, allowing Anthropic to redeploy Claude Mythos 5 to over 100 approved US organizations defending critical infrastructure.
The restriction had also covered Claude Fable 5, which remains suspended for general use, with no confirmed timeline for its return.
Commerce Secretary Howard Lutnick can adjust the approved partner list at any time, signaling continued direct government oversight of the model's distribution.
Mythos 5 has previously been used to uncover thousands of high-severity software vulnerabilities through Anthropic's Project Glasswing partnership network.
The decision reflects a broader trend of US government involvement in approving access to the most capable frontier AI systems, paralleling similar restrictions affecting other AI labs.

Loading Article...

Loading Article...