The US lifted the restriction on two state-of-the-art models from Anthropic
"We received notice that the US Department of Commerce has removed export controls on Claude Fable 5 and Mythos 5. We will begin restoring access from tomorrow," Anthropic said in a Twitter post on June 30.
The announcement was made 4 days after the AI company was allowed by the US government to open access to the Mythos 5 model to a small group of cybersecurity companies in this country.
In a letter dated June 26 and obtained by Reuters, US Secretary of Commerce Howard Lutnick said that Anthropic had coordinated with the US government to address "threats related to the models in question".
OpenAI last week also complied with a request from the government and restricted access to the GPT-5.6 model, allowing only a few approved partners to use it. "This is not a process that we consider optimal," said CEO Sam Altman.
Director of the US Central Intelligence Agency (CIA) John Ratcliffe compared the capabilities of advanced AI models to nuclear weapons, and defended the drastic control measures of President Donald Trump's administration with the launch of the country's most powerful AI technologies.
Anthropic in early June announced Fable 5 and Mythos 5, two models that the company advertises as the most modern by many scales in the AI industry. They are built on the Claude Mythos Preview platform, which once made US officials worried because of its advanced cybersecurity capabilities.
On June 12, Anthropic received instructions from the US government to suspend access to Fable 5 and Mythos 5 for "foreign citizens, whether in the US or other countries, including Anthropic employees". This AI startup said that the US government cited "national security" reasons, but did not provide details.
David Sacks, co-chair of the US President's Council of Science and Technology Advisors, said "a trusted partner" of the US government and Anthropic tested Fable and discovered a jailbreak method to bypass the security barrier that separates this commercial model from Mythos' unlimited cybersecurity capabilities.
The US government then asked Anthropic to patch or recall Fable 5 until it was resolved, but CEO Dario Amodei reportedly refused. "The government then had to reluctantly issue export control instructions. They were ready to lift restrictions as soon as the problem was resolved," he said.
Sacks criticized Anthropic's actions as inconsistent with the position of a laboratory that always puts safety first, and has called for Mythos to be regulated as a cybersecurity weapon.
He said the decision to restrict was not related to previous conflicts between Anthropic and the government, adding that Washington always valued this AI startup's technology and considered the problem "very easy to solve".
Diep Anh (According to AFP, CNBC)