What happened
On June 30, the U.S. Department of Commerce lifted the export controls it had placed on Anthropic's two most capable AI models, Claude Fable 5 and Claude Mythos 5. Anthropic said it began restoring access immediately, ending a shutdown that had lasted roughly nineteen days and had cut off customers around the world.
The episode began in early June. Anthropic launched Claude Fable 5 on June 9 as the first publicly available model in its most advanced tier, alongside Claude Mythos 5, a more restricted variant offered to vetted users working in areas such as cybersecurity and the life sciences. Three days later, on June 12, the U.S. government issued an export-control directive, citing national security authorities, that barred access to both models by any foreign national, whether inside or outside the United States — a restriction that even applied to Anthropic's own foreign-national employees.
Because the company had no practical way to verify the nationality of every user in real time, complying with the directive meant switching both models off for everyone. Anthropic disabled Fable 5 and Mythos 5 globally within hours of the order.
Why the government stepped in
Reporting by Fortune, Forbes, and others indicated the directive followed the government becoming aware of a method for bypassing, or "jailbreaking," the safeguards on Fable 5. Anthropic noted publicly that the government had not provided it with a specific reason for the order, and officials described the concern only in general national-security terms.
The reversal came with conditions. In a letter announcing the decision, Commerce Secretary Howard Lutnick said Anthropic must proactively detect and address security risks in its models, work with the government on standards for future frontier releases, and report malicious activity it observes.
An unprecedented intervention
This is the first time the U.S. government has used export-control authorities to switch off a commercial AI model that was already on the market. Export controls have long applied to physical goods such as advanced chips, but extending them to a software service — and doing so after launch, with immediate effect — is new territory.
The disruption was significant. Businesses that had built products and workflows on the affected models lost access with little warning, and coverage of the episode described it as one of the most consequential outages in the short history of enterprise AI. Some observers questioned whether the intervention was proportionate; others saw it as a preview of how governments may treat the most capable AI systems going forward.
Why it matters
For everyday users and organizations, the story carries a few practical lessons:
- AI is being treated as strategic infrastructure. Governments now consider the most capable models a national-security matter, not just a consumer product. That changes how, and how quickly, such models can be released or withdrawn.
- Access can disappear overnight. Organizations that depend on a single AI provider — or a single model — experienced firsthand what a sudden shutdown means. Contingency planning for AI services is no longer hypothetical.
- Safety findings have real consequences. A reported jailbreak was enough to trigger government action. How companies test, disclose, and patch model vulnerabilities is becoming a matter of public policy, not just internal practice.
What to watch next
The conditions attached to the reinstatement suggest a closer, ongoing relationship between frontier AI labs and the U.S. government. Watch for whether similar review arrangements become standard for future releases across the industry — OpenAI has already said it shared its newest models with the government before its latest preview — and whether Congress moves to define clearer rules for when and how such interventions can happen. The open question raised by researchers after the shutdown remains unresolved: does the government now effectively approve every frontier model release?
