Just two weeks after Australia’s national security agencies and critical infrastructure defenders were granted access to Anthropic’s most powerful AI model Mythos, the US government has pulled the plug citing national security concerns.

A directive form the Trump administration has forced Anthropic to restrict access to its Mythos 5 and the less powerful Fable 5 AI models and to all non-Americans, including those residing in the US and employees of the tech company.

Anthropic said it has temporarily blocked access to the systems while it works with US authorities to address concerns.

The move has sparked fresh concerns about Australia's reliance on overseas technology and renewed calls for sovereign AI capability.

Powerful new model

Earlier this year Anthropic unveiled its new Mythos model, which has not been released publicly.

The company said Mythos had found “thousands” of significant cybersecurity vulnerabilities, including some in every major operating system and web browser.

Describing the model as too powerful for public release, Anthropic has instead been providing access to government agencies, major technology companies and critical infrastructure operators to help strengthen cyber defences.

Australia’s national security agencies and critical infrastructure defenders were granted access to the Mythos Preview AI model just two weeks ago.

Last week Anthropic publicly released a different version of Mythos without the powerful cyber capabilities, dubbed Fable 5.

Wake-up call

A number of AI experts said the US government’s move should serve as a wake-up call for Australia and its reliance on AI services owned overseas.

Australian Computer Society (ACS) interim CEO Dr Prins Ralston said it’s important for Australia to build sovereign capability in AI systems.

“Australia will continue to benefit from global AI platforms and trusted international partnerships,” Ralston said.

“But we should also be clear that our long-term goal must be to build stronger AI skills and capability here at home.

“The Anthropic story is a useful reminder that access to advanced AI systems can be shaped by decisions made offshore, including legitimate national security and safety considerations.”

La Trobe University Chief AI Officer Phil Laufenberg said recent events demonstrate how risky it is to not have these sovereign capabilities.

“In the future, a nation’s productivity will depend on access to frontier models,” Laufenberg said on LinkedIn.

“If these can be shut off within hours, it is easy to imagine how this could be used in adversarial situations.

“Any process that relies on these models that could be revoked would break with precisely no warning at all.”

Digital transformation and AI expert Craig Thomler said the restriction of access to Fable and Mythos “lifts the global AI Cold War to a new level”.

“Nations need their own sovereign capability or face the risk of becoming the new ‘have nots’ of the global economy’,” Thomler posted on LinkedIn.

ACS president Beau Tydd said work needs to be done now to uplift AI skills across the Australian economy.

“For technology professionals, this reinforces how important AI capability will be across the workforce,” Tydd said.

“As AI becomes embedded in business, government and critical services, we need professionals who can understand these systems, assess risk, manage dependencies and deploy AI responsibly.

“Global platforms will remain important, and Australia should be building the professional capability to use AI on our own terms, in ways that support our economy, institutions and public trust.”

‘Misunderstanding’

In a statement posted to the Anthropic website, the company said the US government’s concerns were centred on a potential way to bypass the safeguards put in place for Fable 5.

“The letter did not provide specific details of its national security concern,” the Anthropic statement said.

“Our understanding is that the government believes it has become aware of a method of bypassing, or ‘jailbreaking’ Fable 5.”

The company said it had reviewed a demonstration of a technique that was being used to get Fable 5 to identify a “small number of previously known, minor vulnerabilities”.

“These vulnerabilities all appear relatively simple, and we have found that other publicly available models are able to discover them as well without requiring a bypass,” Anthropic said.

Anthropic said no testers of Fable 5 have found a “universal jailbreak”, which would allow for the unlocking of the wider range of cyber capabilities available through Mythos.

Instead, Anthropic said that it had been given verbal evidence from the US administration of a potential “narrow, non-universal jailbreak” of Fable 5, which involved asking the model to read a specific codebase to fix any software flaws.

“We disagree that the finding of a narrow potential jailbreak should be cause for recalling a commercial model deployed to hundreds of millions of people,” Anthropic said.

“If this standard was applied across the industry, we believe it would essentially halt all new model deployments for all frontier model providers.

“We believe this is a misunderstanding and are working to restore access as soon as possible.”

The banning order was confirmed by Pentagon chief information officer Kirsten Davies on X.

“Some things are simply more important than revenue cycles, clickbait and pre-IPO valuation. American First. Always,” Davies posted.

Anthropic was locked in a feud with the US government earlier this year over the potential use of its AI systems in warfare.

The White House has previously criticised Anthropic over its advocacy for AI safeguards and ethical usage.

Earlier this month Anthropic confidentially filed for an IPO, along with its rival OpenAI.

Anthropic has a reported valuation of $US965 billion ($1.365 trillion).