Anthropic confirmed it had restored its Fable 5 and Mythos 5 systems, 19 days after the White House issued an export control order effectively banning them.

The company pledged to work more closely with the US government on testing AI systems before releasing them.

The White House and Anthropic have repeatedly clashed in recent months.

Dario Amodei, the Anthropic boss, has criticised the US president over his approach to AI and his governance style. Mr Trump has hit back by claiming that the company is run by “Left-wing nut jobs”.

The US is also growing increasingly concerned that restricting advanced AI systems could allow China to catch up in the race to dominate the technology.

Chinese AI companies have released systems in recent days that researchers say are closing the gap on their American rivals.

Fable and Mythos belong to a new class of powerful AI systems that Anthropic has said represent new cyber security risks because of their ability to find and exploit software bugs.

They are widely seen as the most advanced systems released to date.

Fable is a restricted version that was made available to the public last month while Mythos is only available to an approved set of large companies which can use it to improve their own security.

The commerce department had intervened in June owing to concerns about a “jailbreak” that it said would have bypassed the system’s safety features and allowed it to be used for cyber attacks.

The export order barred non-US citizens from accessing the two AI systems, leading to concerns that Britain was becoming overdependent on systems that the White House could cut off at a moment’s notice.

Anthropic had responded to the order by blocking all access – including to Americans – since it does not collect data on users’ citizenship.

The US commerce department had raised concerns about the systems when researchers at Amazon discovered a way to bypass safeguards that stop users from carrying out cyber attacks.

Anthropic has downplayed the severity of the jailbreak but said on Tuesday that it had blocked it.

However, it said this meant its systems were now less capable than before. The company added that some requests would be downgraded to less powerful AI models to avoid misuse.

Chinese company Zhipu AI released a system last month that security researchers said matched Mythos when it came to finding security bugs.

In a letter to Anthropic, Howard Lutnick, Mr Trump’s commerce secretary, said: “Anthropic has agreed to proactively detect and address security risks associated with the models; to work diligently with the US government… and to inform the US government of any malicious activity.”

The letter was addressed to Tom Brown, Anthropic’s chief compute officer, rather than Mr Amodei. Mr Brown has led discussions with the White House in recent days.

Anthropic’s main rival OpenAI agreed last week to limit access to its latest model, GPT 5.6, at the request of the US government.

Source link