US Clears Anthropic's Mythos 5 for Limited Deployment in Landmark AI Security Shift

Washington | EcoPulse24

The United States has authorized Anthropic to restore limited access to its advanced Mythos 5 artificial intelligence model after the company addressed national security concerns raised by the Trump administration, marking what could become a defining moment in how governments oversee the deployment of frontier AI systems.

Under the decision, Anthropic will be permitted to redeploy Mythos 5 to a select group of trusted cybersecurity and infrastructure partners, while restrictions on its related Fable 5 model remain in place pending further discussions with US officials.

The approval follows nearly two weeks of negotiations between Anthropic and the US Department of Commerce after Washington abruptly suspended foreign access to both models over concerns that their security safeguards could potentially be bypassed.

Commerce Department Eases Restrictions on Mythos 5

In a letter reviewed by Bloomberg News, US Commerce Secretary Howard Lutnick said Anthropic had made "significant progress" in addressing risks associated with the advanced AI models.

"These efforts have yielded significant progress," Lutnick wrote, adding that Mythos 5 could now be released to "certain trusted partners."

Anthropic confirmed it had received authorization to restore access to its flagship cybersecurity model and said deployment would begin immediately for the approved organizations.

"We are pleased to see this progress and continue to work with the government to expand access to Mythos 5 and make Fable 5 available for general use again," the company said in a statement.

Fable 5 Remains Under Government Restrictions

While Mythos 5 has been partially cleared for redeployment, the Commerce Department did not announce any changes regarding Fable 5, indicating that US authorities remain cautious about wider public access to the company's most advanced AI capabilities.

Anthropic has not disclosed what technical measures it implemented to satisfy government concerns surrounding potential jailbreaking, or attempts to bypass the models' built-in security guardrails.

The company said discussions with federal officials are continuing as it seeks broader approval for both models.

OpenAI Also Adjusted GPT-5.6 Rollout

The Anthropic decision comes as OpenAI has also modified its deployment strategy for GPT-5.6, announcing that an initial preview version will be made available only to a limited group of partners approved by the US government before expanding access more broadly.

The parallel developments suggest that leading AI developers are increasingly coordinating frontier model releases with US regulators, particularly for systems possessing advanced cybersecurity or dual-use capabilities.

Project Glasswing Includes Leading Technology Firms

Before the restrictions were imposed, Mythos 5 had been deployed through Project Glasswing, an invitation-only initiative serving approximately 200 organizations specializing in cybersecurity and critical infrastructure.

Participants include major technology companies such as Apple, Microsoft, Google, Cisco, and Nvidia, highlighting the model's strategic role in identifying cybersecurity vulnerabilities and strengthening digital infrastructure.

AI Oversight Becomes a National Security Issue

The temporary suspension of Anthropic's models represented one of the most significant government interventions to date in the commercial deployment of frontier artificial intelligence.

The episode also unfolded as Anthropic moves toward a potential initial public offering after confidentially filing IPO documents, with Bloomberg previously reporting the company's valuation has surpassed $900 billion.

EcoPulse24 Analysis

The significance of Washington's decision extends well beyond Anthropic.

Rather than permanently restricting one of the world's most advanced AI models, the US government has introduced what increasingly resembles a controlled-access framework for frontier artificial intelligence - allowing deployment only to approved organizations while maintaining tighter restrictions on broader public availability.

The decision also appears to establish an emerging regulatory precedent. Anthropic's negotiations with the Commerce Department, combined with OpenAI's decision to limit the initial release of GPT-5.6 to government-approved partners, suggest that regulatory clearance may become an integral part of future frontier AI launches alongside technical readiness.

For governments investing heavily in sovereign AI capabilities - including the United Arab Emirates and Saudi Arabia - the development reinforces a broader transformation in which advanced AI models are increasingly treated as strategic national assets, subject not only to commercial considerations but also to national security oversight.

If this framework continues to evolve, future competition in artificial intelligence may be shaped not only by computational power and model performance, but also by governments' willingness to determine who can access the world's most capable AI systems, under what conditions, and for what purposes.