Anthropic restores Fable AI model after US lifts export restrictions

The US Commerce Department has lifted export controls on Anthropic's Fable and Mythos artificial intelligence models, reversing restrictions imposed earlier this month after the government raised national security concerns over the company's most advanced AI systems.Anthropic confirmed the decision on Tuesday, saying it would begin restoring access to both models after receiving formal notice from the Commerce Department. The restrictions had been introduced on June 12, forcing the AI company to suspend access to its Fable 5 and Mythos 5 models while the government assessed potential security risks.Last week, the department partially eased the restrictions by permitting Anthropic to make Mythos 5 available only to a limited group of trusted US organisations. The latest decision removes the broader export controls, allowing the company to resume wider access.According to a letter from US Commerce Secretary Howard Lutnick, Anthropic committed to strengthening its security oversight in exchange for the restrictions being withdrawn. The company agreed to proactively identify and address security risks associated with its AI models, work closely with the US government on safety protocols for current and future releases, and report malicious activity involving its systems.However, the Commerce Department said it retains the authority to reinstate licensing requirements if circumstances change or if Anthropic fails to meet its commitments.Fable and Mythos are built on the same underlying AI architecture but are designed for different use cases. While Fable is intended for broader public deployment, Mythos provides fewer built-in safeguards and is aimed at more advanced applications.Anthropic also introduced an additional safeguard designed to prevent a technique that regulators believed could be used to bypass protections built into Fable. A company source said the security vulnerabilities associated with that method had already been identified and addressed before the government's intervention.The restrictions reflected growing US scrutiny of frontier AI models amid concerns that increasingly capable systems could be exploited for cyberattacks or by foreign military and intelligence agencies. Security experts have warned that advanced AI models with fewer operational constraints could accelerate sophisticated cyber operations, particularly against sectors with complex digital infrastructure such as financial services.The government's decision follows a broader push to strengthen oversight of advanced AI development. Earlier this month, President Donald Trump signed an executive order establishing a voluntary framework that encourages developers of frontier AI models to provide the US government with early access to certain systems before broader deployment.The move also comes after OpenAI delayed the full public rollout of its GPT-5.6 model at the request of the US government, limiting initial access to a select group of vetted partners.Also read: Anthropic launches Claude Science AI platform for researchers, expands into drug discovery