The Trump administration lifted export controls on Anthropic's Claude Fable 5 on Tuesday, Commerce Secretary Howard Lutnick said in a post on X, and Anthropic said it will begin restoring access to Claude Fable 5 and Mythos 5 on Wednesday; Mythos 5 was allowed back last week to around 100 trusted organizations.
Anthropic released Claude Fable 5 to the public on June 9 with safeguards the company said would reduce the risk of misuse, and said Mythos 5 — a version with fewer guardrails — would initially be made available to a select group of major companies for testing purposes.
On June 12 the models were forced offline after senior administration officials said the systems posed severe cybersecurity risks and Lutnick enacted export controls preventing any foreign national from accessing the models, a move that required Anthropic to disable the products.
Anthropic dispatched a team of its top scientists to Washington to hammer out a solution with government officials after the takedown.
Lutnick said Anthropic’s cooperation with government officials was a key reason for the detente with the company.
Lutnick sent a letter Tuesday informing Anthropic co-founder Tom Brown of the decision and said Anthropic agreed to continue collaborating "on protocols and standards and releases for Mythos, Fable, and future models" and "to inform the U.S. government of any malicious activity."
The Commerce Department said in a letter that Anthropic has addressed the risks. "Anthropic has agreed to proactively detect and address security risks associated with the models," Lutnick wrote, and the department said it reserves the right to reconsider the decision to lift export restrictions if necessary.
Fable 5 included guardrails that prevented the system from assisting with a wide range of cybersecurity- and biology-related tasks, measures the company said it put in place because of bad actors who use powerful AI systems for malicious cyber and biology purposes.
Anthropic thanked its users on X and said it would soon share a further update.
The company said the government's concerns appeared focused on a potential "jailbreak" technique and called the vulnerabilities found using the apparent jailbreak "relatively simple" and replicable with other AI models, adding, "[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." Experts disagreed over the severity of the risk.
Fable 5 is a version of the AI model for the consumer market capable of deep reasoning and performing complex tasks independently, while Mythos 5 is a version designed for businesses and cybersecurity experts and is said to be able to identify vulnerabilities in computer code and exploit them; both models were released on June 9.
President Trump signed an executive order earlier this month creating a voluntary 30-day review process for the most advanced private AI models; Mr. Trump had delayed signing the order, saying he did not want to "get in the way" of U.S. leadership of the AI industry or risk giving China an edge. White House chief of staff Susie Wiles thanked companies that had cooperated with the order on "advanced model access and guardrail testing and security."
Anthropic earlier clashed with the Pentagon over the company's push for formal guardrails to prevent the military from using its Claude models to power fully autonomous weapons or mass surveillance. After the two sides failed to reach an agreement, Mr. Trump ordered all federal agencies to stop using Anthropic's AI technology and Defense Secretary Pete Hegseth labeled it a "supply chain risk." Anthropic sued, and a federal judge blocked those restrictions, calling them "Orwellian" and an effort to "cripple" the company; the federal government is appealing.
OpenAI rolled out GPT-5.6 only to a small set of approved customers last week after a request from the U.S. government.
U.S. officials, allies and leading AI companies have grown increasingly concerned that frontier models could be misused to automate sophisticated cyberattacks or accelerate biological weapons development.
China is moving closer to producing its own Mythos competitor while continuing to release open-weight models that rival models like Anthropic's Claude Opus 4.8.
The administration faces an August deadline under a recent executive order to create standardized benchmarks for evaluating the security risks of new AI models.