The Trump administration has ended its ban on two of Anthropic’s most advanced artificial intelligence models, wrapping up a weeks-long restriction that had been put in place over cybersecurity worries.
The San Francisco-based AI company announced Tuesday evening that its model known as Claude Fable 5 is now available to the general public. A more powerful model called Mythos 5 is also being restored, though access will be limited to a hand-picked group of U.S.-based organizations that have received federal government approval.
The restrictions began on June 12, when the Commerce Department blocked foreign nationals from using either AI model. Anthropic said the move forced the company to pull both products entirely — for all users worldwide — just days after they had been introduced to the public.
In a blog post this week, Anthropic explained that the government’s alarm was set off by a report from cybersecurity researchers at Amazon, which serves as Anthropic’s main cloud computing partner. According to Anthropic, those researchers “had found a method of bypassing Fable 5’s safeguards” that could allow someone to identify and potentially take advantage of weaknesses in software systems.
Concern had already been building earlier this year after Anthropic flagged that its Mythos model showed a troubling ability to detect software flaws — a capability that, in the wrong hands, could be used by hackers to attack critical computer networks across the globe.
Anthropic’s top competitor, OpenAI — the company behind ChatGPT — also announced Friday that it is holding back the release of its newest AI model at the request of President Donald Trump’s administration. OpenAI said its new product, called GPT-5.6 Sol, will only be available to a limited group of government-approved customers for a temporary period.
Last month, President Trump signed an executive order establishing a system for the federal government to review the national security risks posed by the most advanced AI systems before they reach the public — allowing up to 30 days for that review process. While the order describes AI developers’ participation as voluntary, the full framework for carrying it out has not yet been finalized.
