Anthropic is engaged in urgent talks with Trump administration officials in Washington on Monday in an attempt to resolve a new dispute over its most advanced AI models, following a surprise export control directive that has temporarily halted access to Fable 5 and Mythos 5 for all users.
The directive, issued Friday under national security authorities, ordered the company to suspend access to the models “by any foreign national, whether inside or outside the United States.” In response, Anthropic disabled the models entirely to ensure full compliance, a move that affects hundreds of thousands of enterprise customers and paid subscribers who had just begun using the newly released systems.
The action marks the latest flashpoint in Anthropic’s increasingly complicated relationship with the U.S. government. Earlier this year, the Pentagon labeled the company a supply chain risk after it refused to allow its Claude models to be used for domestic surveillance or fully autonomous weapons. That designation banned defense contractors from using Anthropic technology and prompted the startup to file a lawsuit that remains ongoing.
Register for Tekedia Mini-MBA edition 20 (June 8 – Sept 5, 2026).
Register for Tekedia AI in Business Masterclass.
Join Tekedia Capital Syndicate and co-invest in great global startups.
Register for Tekedia AI Lab.
Defense Secretary Pete Hegseth defended the latest restrictions on X, writing that “every passing day” proves why blacklisting Anthropic was “the right move.”
Fable 5 and Mythos 5 were unveiled just days before the directive. They build on Anthropic’s earlier Claude Mythos Preview, which demonstrated strong capabilities in identifying software vulnerabilities as part of the company’s Project Glasswing cybersecurity initiative. The new models were tested with government agencies prior to release and received initial approval for deployment, according to a person familiar with the process.
Anthropic described the government’s sudden concern as focused on a “potential narrow, non-universal jailbreak,” where a user might bypass safeguards to ask the model to analyze and fix flaws in a specific codebase.
But the company has pushed back strongly in a statement. It said: “We disagree that the finding of a narrow potential jailbreak should be cause for recalling a commercial model deployed to hundreds of millions of people. If this standard was applied across the industry, we believe it would essentially halt all new model deployments for all frontier model providers.”
Anthropic characterized the dispute as a “misunderstanding” and said it is working to restore access “as soon as possible.”
Cybersecurity Community Pushes Back
The restrictions have drawn criticism from cybersecurity leaders across major U.S. firms. In a letter sent Sunday and signed by more than 50 security experts from companies including Nvidia and Adobe, the group argued that removing access to Anthropic’s models hampers defenders at a time when other AI tools are making it easier for attackers to find and exploit vulnerabilities.
The letter emphasized that Anthropic’s models are not uniquely dangerous and that China is advancing rapidly in its own AI capabilities. Taking the best defensive tools off the table, they warned, is “dangerous” for U.S. cybersecurity leadership.
This echoes Anthropic’s own position that the action creates market uncertainty and risks America’s AI edge without clear justification. CrowdStrike recently identified China-linked hackers as the biggest espionage threat to technology companies over the past year, adding weight to concerns about ceding ground in defensive AI.
Anthropic is now walking on a tightrope for a company that has positioned itself as a safety-focused leader, refusing certain military applications and limiting initial rollouts of powerful models like Mythos. As it scales commercially, with a valuation nearing $965 billion and a confidential IPO filing underway, it faces growing pressure to balance ethical commitments with national security expectations.
The Trump administration’s approach appears to reflect broader concerns about dual-use AI technologies that could be repurposed for offensive cyber operations. At the same time, the swift backlash from the cybersecurity community underscores fears that overly broad restrictions could weaken America’s overall defensive posture against sophisticated state actors.
Analysts note that Anthropic’s willingness to engage directly with officials on Monday suggests both sides are motivated to find a resolution. A productive outcome could ease tensions and allow the models back into circulation with enhanced safeguards. Failure to resolve the issue quickly, however, risks further straining relationships at a time when U.S. AI leadership is seen as strategically vital.
However, the dispute is largely seen as a microcosm of larger challenges facing the AI sector. As models grow more capable, the line between beneficial tools and potential national security risks becomes harder to draw. Regulators and companies are still developing frameworks for responsible deployment, and cases like this will help shape how those frameworks evolve.