The US Department of Commerce has reversed a controversial export ban on AI giant Anthropic’s frontier models Claude Fable and the more cybersecurity-focused Claude Mythos – though Australian agencies have not regained accessed to the latter.

In mid-June, the Trump administration mandated that Anthropic restrict non-Americans from using Fable and the unreleased Mythos, including those residing in the US.

Less than three weeks later, the ban has been nixed following a period of consultation with Anthropic.

“As of today… the export controls on Fable 5 and Mythos 5 have been lifted,” Anthropic wrote Wednesday.

Fable has been redeployed for public use as of Thursday (AEDT), though it’s understood Mythos has only been restored for a specific set of US organisations.

“We continue to coordinate with the [US] government to expand access to the broader set of domestic and international partners in the Glasswing [Mythos preview] program,” Anthropic said.

The company was contacted for comment on when Australian partners may regain access to Mythos.

The US export ban came only two weeks after limited organisations, including allied Australian security agencies, were granted access to a preview model of Mythos.

At the time of the ban, an Australian Signals Directorate spokesperson told Information Age it was working with US and other international partners on the “implications of frontier AI models”.

“We remain committed to continuing our longstanding collaboration on cybersecurity with the US government,” they said.

The White House initially blocked access to Fable and Mythos due to apparent risks to US national security, though a letter to Anthropic confirmed the expert controls have been withdrawn.

“Anthropic has agreed to proactively detect and address security ​risks associated with the models,” US commerce secretary Howard Lutnick said in a letter seen by the BBC.

Lutnick reportedly added, however, that the Department of Commerce reserved the right to “reevaluate the decisions made in this letter and ​the necessity of reimposing a license requirement”.

The export controls came after Anthropic successfully marketed Mythos as an unprecedented frontier model that could unearth cybersecurity vulnerabilities in the world’s most sensitive and popular software.

Ban came after Fable safeguards cracked

Notably, the Trump administration’s export controls came shortly after Anthropic’s safety controls for Fable 5 were allegedly ‘jailbroken’ by X user ‘Pliny the Liberator’.

Though Fable sports the same underlying ‘brain’ as Mythos, it was released for public use thanks to a series of safeguards that divert potentially dangerous questions to other, less capable models.

Pliny claimed to have cracked these safeguards on 11 June, about two days after the model was released.

Anthropic, meanwhile, said the export control directive came on 12 June after the US government became aware of a safety bypass method discovered by researchers at Amazon.

“In one case, the model produced code demonstrating how [a] relevant vulnerability could be exploited,” wrote Anthropic.

After working with the US government and Amazon, Anthropic ultimately determined this “reported technique did not expose any unique Mythos-level cyber capabilities”, and further bolstered Fable’s safeguards.

Call for AI sovereignty as US dangles frontier models

Niusha Shafiabady, director of the Women in AI for Social Good lab at Australian Catholic University, said although Australia should welcome the easing of restrictions, we shouldn’t see it as a “complete solution”.

“Access to overseas AI models is important, but it is not the same as AI sovereignty,” Shafiabady told Information Age.

“The real opportunity for Australia is to build and support local AI capability, with ethical and responsible AI principles embedded from the beginning.

“That means models designed around Australian values, Australian regulations, local industry needs, and public trust.”

Shafiabady said the global AI landscape had become “uncertain”.

“A model like Fable 5 can be restricted one week and cleared for export the next,” she said.

“That does not necessarily mean the risk has disappeared – it means the risk is being managed differently.”

Minister for Industry and Innovation, Tim Ayres, and Assistant Minister for Science, Technology, and the Digital Economy, Andrew Charlton, were contacted for comment.

Anthropic rival OpenAI said in late June that it would delay a full public launch of its next-generation GPT-5.6 model at the US government’s request, but added, “We don’t believe this kind of government access process should become the long-term default.”