The US government has partially lifted its unprecedented ban on Anthropic’s Claude Mythos 5, allowing roughly 100 vetted American companies and federal agencies to access the frontier cybersecurity AI model under new safeguards. The decision, confirmed by Commerce Secretary Howard Lutnick on June 26, ends a two-week freeze that represented the most significant federal intervention into an AI company’s operations to date.
The Ban That Shook the Industry
On June 12, just three days after Anthropic launched Fable 5 and Mythos 5 to the public, the Commerce Department ordered the company to disable both models within 90 minutes. The directive, issued under export control authority, cited national security concerns around the models’ advanced cybersecurity and biological capabilities. Anthropic had previously granted early access to approximately 40 organizations responsible for critical infrastructure — allowing them to patch vulnerabilities before malicious actors could exploit them — but the blanket ban shut off all access worldwide.
Two Weeks of Diplomacy
During the blackout, Anthropic dispatched senior members from its cybersecurity and AI safety teams to Washington, DC to negotiate directly with Trump administration officials. The company argued that Mythos 5’s cybersecurity capabilities were designed to help defend critical infrastructure, not threaten it. Those negotiations culminated in Friday’s partial restoration agreement.
Under the deal, Anthropic agreed to implement additional safeguards on its models. The restored access is limited to “trusted” US organizations — a list that includes large corporations, government agencies, and infrastructure operators. The model remains unavailable to foreign nationals and international customers.
Fable 5 Still Restricted
While Mythos 5 got the green light, its sibling model Fable 5 remains under tighter controls. Fable 5 is now explicitly prevented from responding to certain categories of queries, including those related to cybersecurity and biology. In those cases, Anthropic routes requests through Claude Opus 4.8 instead — a model with more constrained capabilities in those domains.
What This Means for AI Governance
The Mythos 5 episode establishes a new precedent: frontier AI models can now face government-mandated shutdowns on national security grounds, with negotiated safequards as the path back to deployment. The simultaneous June 26-27 announcements of GPT-5.6’s government-coordinated preview and Mythos 5’s partial restoration signal what observers are calling “the new pattern for frontier AI releases in the US” — one where Washington has a direct seat at the deployment table.
In related Anthropic news, the company also raised API rate limits across its Claude developer platform on June 26, bringing Sonnet and Haiku rate limits to parity with Opus at every usage tier — a welcome change for developers building agentic applications on the Claude API.