US Releases Restrictions on Anthropic’s AI Mythos Model
Antropic and the Trump administration reached an agreement Friday to bring one of the company’s most powerful AI models back online, after days of tense negotiations.
In a letter sent to Anthropic, the Commerce Department granted the San Francisco company permission to restore some clients’ access to its Mythos 5 model, which the government restricted two weeks ago due to national security concerns.
Talks were still ongoing about restoring access to another powerful anthropic model, Fable 5, and giving Mythos to more organizations, said a person familiar with the talks, who spoke on condition of anonymity because the person was not authorized to discuss them publicly.
“Anthropic has been working with the US government to address the risks associated with covered models,” said a letter Commerce Secretary Howard Lutnick sent to Anthropic on Friday. “These efforts have produced significant progress.”
Mr. Lutnick added that Anthropic is committed to working with the US government on protocols for future releases of AI models.
A spokesperson for Anthropic said the company is working to restore access to administration-approved clients and is “pleased to see this progress and continue to work with the government.”
He reported on the letter earlier Traffic lights.
The deal de-escalates a dispute between the Trump administration and Anthropic, one of the most influential AI companies. But it may do little to assuage concerns that the administration intends to take a more active role in issuing new AI systems.
As part of the deal, Anthropic pledged to implement safeguards on its AI models in exchange for restoring access to hundreds of clients. On June 12, the company’s sales department suddenly said it had 90 minutes to deactivate Mythos and Fable.
Anthropic later learned that the administration’s concerns stemmed from a paper published by Amazon researchers that pointed to a perceived security flaw. The researchers were able to convince the AI model to detect bugs in pieces of vulnerable software code.
Amazon did not include tests of models from other AI firms, some of which are capable of generating the same information, said cybersecurity experts who have seen or been briefed on the report.
It was unclear whether the administration would lift restrictions on Anthropic entirely or create a more formal process that other AI companies could follow to avoid similar problems.
Anthropic and the Trump administration first clashed earlier this year over a $200 million Pentagon AI contract
The company and the Pentagon disagreed on how the technology should be used in war. This led Defense Secretary Pete Hegseth to label the Anthropocene a “supply chain risk” in February. The designation, which meant the company posed a national security risk, had never been used against a US company. In March, Anthropic sued the government over the label.
Anthropic released Mythos to a small group of companies and researchers in April. The artificial intelligence company said its new AI model was particularly effective at finding vulnerabilities in software, raising concerns among security researchers and government officials that it could be used in cyberattacks.
Anthropic followed up on Fable, which he said had so-called gatekeepers that would limit what users could do with it. However, the researchers speculated that these barriers could be easily breached, as was the case with most other AI systems.
When the government ordered Anthropica to take its new models offline, some at the company questioned whether the company was being unfairly singled out. However, the administration’s interest in managing the release of AI models appears to be expanding.
Earlier on Friday, OpenAI, Anthropic’s main rival, unveiled an AI technology called GPT-5.6 Sol and said it matched Mythos’ performance in standard benchmark tests used by AI companies. OpenAI said it would initially share the technology only with a small number of companies approved by the administration.
“We do not believe that this kind of government access process should become the long-term standard,” the company said blog post. “It keeps the best tools in front of the users, developers, enterprises, cyber defenders and global partners who need them. We’re taking this short-term step because we believe it’s the strongest path to wider availability.”
(The New York Times has sued OpenAI and Microsoft, alleging copyright infringement of news content related to AI systems. Both companies have denied the claims.)
The administration also pressed Meta this week to submit its AI models to voluntary review.
Meta is the only major US AI developer that has not reached an agreement to voluntarily share its models with the federal government for review. OpenAI, Anthropic, Google, xAI and Microsoft have all agreed to submit their models to the government’s AI safety group, known as the Center for AI Standards and Innovation.
The center, overseen by Mr. Lutnick, was created by the Biden administration to vet artificial intelligence models and has a technical staff that conducts those evaluations.
Cade Metz, Tripp Mickle and Tyler Pager contributed reporting.