US orders Anthropic to halt foreign access to its most advanced AI models

Artificial intelligence company Anthropic has taken its most advanced AI models offline after it was ordered by the US government to suspend access for all foreign nationals.

This comes after the company recently gave Australian companies and individuals access to these new models, which it initially said were "too dangerous" for public release.

On Saturday morning, Australian time, Anthropic published a blog confirming reports that it had received an export control directive from the Trump administration to suspend access for foreign nationals to its Fable 5 and Mythos 5 AI models.

In a statement, Anthropic said the US government has not provided details for its "national security concern", but it understands the government believes it has become aware of a method of bypassing its models' safeguards.

The company voluntarily places guardrails around the use of its AI models to thwart their use in cyber attacks, the creation of bioweapons and other harmful activities.

"The net effect of this order is that we must abruptly disable Fable 5 and Mythos 5 for all our customers to ensure compliance," Anthropic said.

"Access to all other Anthropic models will not be affected."

Anthropic added that it believed there was a "misunderstanding" and that it was working to restore access to the models as soon as possible.

The geopolitics of access to AI

The sudden shutdown lands in the middle of a broader fight over who gets access to the world's most powerful AI systems.

Last month, the ABC reported that access to frontier models had become part of Australia's discussions with Anthropic and other major technology companies, as the federal government sought to lure more AI investment into the country.

Earlier this month, an Anthropic spokeswoman confirmed Australian organisations would be included in an expanded Mythos access program, although the company declined to name specific companies or government agencies.

Australian AI start-up monō ai founder David Hyman said he was using one of Anthropic's new models when it got pulled offline.

"Was mid-build […] and got an error I've never seen before. 'Model not found'," he said.

This week, Anthropic chief executive officer Dario Amodei argued that governments should have stronger powers over the release of dangerous AI systems.

In a blog post, Mr Amodei said frontier AI models should be subject to technical testing and auditing, and that governments should be able to block or reverse the release of a model if it presented unacceptable public safety risks.

The Trump administration has already been in conflict with Anthropic.

In March, the Department of War confirmed the company had been designated a supply-chain risk to America's national security, a decision the company is challenging in court.

Australian Strategic Policy Institute senior analyst Fergus Ryan said that Australia should still get access to models that the US wishes to restrict for national security reasons.

"If Washington now sees frontier AI models as strategic capabilities, it needs an allied-access regime for Five Eyes partners, not a one-size-fits-all foreign exclusion rule," he said.

Department of Home Affairs and the office of Minister Tony Bourke have been contacted for comment.

OpenAI under investigation

The ChatGPT maker was served with a subpoena on Friday local time seeking documents related to a wide range of its activities and the impact on users, including advertising, user engagement and retention, and the handling of consumer and health data, the source said.

The subpoena, sent by New York's attorney general, also seeks information on activities related to minors and seniors, deep learning models and internal company policies, the source added.

The probe represents the latest legal challenge for IPO-bound OpenAI, which is being sued by Florida for allegedly misrepresenting the safety of its ChatGPT platform.

The source declined to be identified while discussing the investigation, which has not been publicly announced.

An OpenAI spokesperson said: "AI is a new and powerful technology, and we work every day to safely bring its benefits to people in a responsible way.

"We take the concerns raised by state attorneys general seriously and intend to engage constructively with their offices."

The Florida lawsuit, the first by a US state, claims the platform has harmed children by providing information to school shooters, offering guidance on self-harm and addicting young users.

A Canadian mother also sued OpenAI and Chief Executive Sam Altman in US court on Thursday, alleging ChatGPT encouraged her daughter to kill herself.

OpenAI said last week it had confidentially filed for a US IPO that a source said could come as early as September and value the company at up to $US1 trillion ($1.4 trillion).

ABC/Wires