Anthropic has won approval from the United States government to bring part of its powerful Mythos 5 artificial intelligence model back online.
The decision comes after the company worked with officials to address concerns about national security risks linked to the technology.
According to Bloomberg, US Commerce Secretary Howard Lutnick told Anthropic that the company had made “significant progress” in addressing risks associated with its covered AI models.
The approval means Mythos 5 can now be released again to “certain trusted partners”.
But the move does not fully end the dispute.
The US government has not lifted restrictions on Fable 5, another advanced model linked to Mythos 5.
Why the model was restricted
The confrontation began two weeks ago when the US government barred Anthropic from giving foreign nationals access to Mythos 5 and Fable 5.
Officials were worried that the models’ safety systems could be bypassed through “jailbreaking”.
That order forced Anthropic to disable global access to both systems.
Anthropic said Mythos 5, which it described as its strongest cybersecurity model, can now be redeployed to a small group of cyber defenders and infrastructure providers.
The company said it is working to restore access to approved providers as quickly as possible.
It also said it wants to continue talks with the government to expand access to Mythos 5 and bring Fable 5 back into general use.
Fable 5 remains under pressure
Fable 5 was released as a public-facing version of Anthropic’s Mythos-class model.
But it came with guardrails meant to limit its cyber capabilities.
Anthropic has said Fable 5 is blocked from answering some queries, including certain cybersecurity and biology-related questions.
In those cases, its Claude chatbot routes users to a different model called Opus 4.8.
The US government’s continued restriction of Fable 5 shows officials are still cautious about making the technology widely available.
AI industry feels the impact
The restrictions marked one of the strongest US government interventions yet into the operations of an AI company.
They also sent a message across the wider AI industry.
Bloomberg reported that OpenAI also limited the release of a new model, GPT-5.6, under pressure from the Trump administration.
Anthropic has pushed back against the standard being applied to frontier AI models, warning that “perfect jailbreak resistance” is not currently possible for any provider.
The company is now working with the US government on a policy framework for similar cases in future.