The U.S. Commerce Department has lifted export controls on Anthropic’s advanced Claude Fable 5 and Claude Mythos 5 models, reversing restrictions imposed less than three weeks earlier over national security concerns.
Anthropic said it received notice that the controls had been removed and would begin restoring access to the models. The decision follows a June 12 government directive that forced the company to suspend access to Fable 5 and Mythos 5 for foreign nationals, including foreign-national employees, whether inside or outside the United States. Because Anthropic said it could not reliably verify user nationality in real time, it disabled access more broadly to ensure compliance.
The restrictions were tied to concerns that the models’ advanced cybersecurity capabilities could be misused, particularly if foreign military or intelligence actors obtained access. Anthropic said the government’s action followed a reported method for bypassing some Fable 5 safeguards, allowing the model to identify software vulnerabilities and, in one case, produce code demonstrating how a vulnerability could be exploited.
Fable 5 is Anthropic’s most capable broadly released model, designed for software engineering, knowledge work, research, vision and complex long-horizon tasks. Mythos 5 uses the same underlying model but has some safeguards lifted for approved cybersecurity partners, making it more powerful for vulnerability discovery and defensive security work. Anthropic initially launched Mythos 5 through Project Glasswing in collaboration with the U.S. government.
Safeguards Unlock Model Access
Anthropic said access was restored after it implemented additional safeguards to block the behavior described in the report that triggered the government’s concerns. When certain requests are blocked, they may be routed to the company’s less capable Claude Opus 4.8 model instead. The company acknowledged that the approach may frustrate some users but said it was necessary to make Fable 5’s broader capabilities available safely.
The company also said it is working with Amazon, Microsoft, Google and other partners to develop common standards for assessing and fixing AI jailbreaks. That effort includes a proposed system for ranking jailbreak severity, an important step as frontier AI companies face growing pressure to show that safety decisions are based on measurable technical risk rather than opaque political judgment.
The Commerce Department’s reversal appears conditional. Reuters reported that Commerce Secretary Howard Lutnick told Anthropic the department reserved the right to reevaluate the decision and reimpose licensing requirements if circumstances change or if Anthropic fails to meet its commitments.
Frontier AI Enters Export-Control Era
The episode marks a major escalation in U.S. oversight of frontier AI models. Export controls have historically focused on chips, semiconductor tools and defense-sensitive technologies. Applying similar restrictions to access for specific AI models shows that Washington increasingly views model capability itself as a strategic asset.
The policy implications are significant. AI companies may now need to plan for government review, access limits and post-release monitoring when launching systems with advanced cyber, biological, scientific or autonomous-agent capabilities. Anthropic’s case also shows how difficult enforcement can be when cloud-based models are distributed globally and users interact through APIs, enterprise accounts and consumer interfaces.
For Anthropic, the reversal limits immediate commercial damage, but it also deepens the company’s dependence on government trust. The company said it will provide designated U.S. government partners expanded early access to both models and work with officials on protocols for current and future systems.
The broader market impact extends beyond Anthropic. OpenAI and other frontier labs are also facing closer scrutiny around model deployment, national security review and access for foreign users. The lifting of controls on Fable 5 and Mythos 5 does not end that trend. It instead shows the likely shape of the next phase: powerful AI models can reach global markets, but only with safeguards, reporting obligations and a much tighter relationship between developers and the U.S. government.
