US forces Anthropic to shut down latest AI models, citing national security concerns
AI company Anthropic has disabled access to its latest Mythos and Fable models at the order of the Trump administration, which said it feared they could be deployed to intelligence services in countries like China and Russia and used to commit cyberattacks.
The concern was raised in a letter sent to Anthropic CEO Dario Amodei by US commerce secretary Howard Lutnick on 12 June. According to international news agency Reuters, which has seen a copy of the letter, Lutnick ordered the company to suspend the export of its Claude Mythos 5 and Claude Fable 5 AI models and to prevent its use by foreign nationals, whether they reside in the US or abroad.
Anthropic said the only way it could satisfy the export controls imposed on it was to shut the models down entirely. The models represent the most advanced AI capabilities Anthropic has developed, and had been launched just three days before the US government decided to take action.
According to Anthropic, the government said it believes there is a way of bypassing, or ‘jailbreaking’, a safeguard designed to prevent Fable from being used to identify software vulnerabilities for the purpose of committing cyberattacks.
Anthropic said it disagreed that the jailbreak it was shown warranted the restrictions placed on it, claiming that the bypass found only “minor” security flaws that other publicly available AI models can also find.
“We believe the government should have the ability to block unsafe deployments, as part of a statutory process that is transparent, fair, clear and grounded in technical facts,” Anthropic said. “This action does not adhere to those principles.”
It added that such action could “essentially halt” new frontier AI models if applied broadly.
It is understood that Anthropic staff are meeting with representatives of the Trump administration to negotiate a way forward.
Read more: On the defensive: strengthening government cybersecurity in a changing landscape
US government seeks to review AI models before public release
Prior to the latest action, US president Donald Trump signed an executive order which asks AI developers to voluntarily provide the federal government with access to certain frontier models for a period of up to 30 days prior to releasing them to other organisations, so that it can review them for cybersecurity risks.
It is understood the executive order – which was signed on 2 June and contains several other measures related to AI and cybersecurity – was drawn up following concern about the cybersecurity risks posed by Anthropic models.
“Advanced AI capabilities make our nation stronger, but also introduce new national security considerations that require coordinated action across executive departments and agencies,” the order states.
“As these capabilities evolve, [the administration] will continue to work closely with industry to ensure that the best and most secure technology is deployed rapidly to confront any and all threats to our country.”
It says the administration is taking an approach to cybersecurity innovation that “enhances both our national security and our global AI dominance”.
Taking to the social media platform X to clarify the scope of the order, venture capitalist and former White House advisor, David Sacks, said that the government would only review “models that represent a meaningful step-change in cyber capabilities”, rather than “incremental version numbers of existing models”.
He added that the review period had been changed from 90 days prior to wider release to 30 days, and called the move “a game changer” that would allow the country’s AI labs to “comply with the voluntary framework without delaying new model releases”.
“In the AI race, every day counts,” he said.
Read more: Lessons from Canadian cyber-defence
Cybersecurity ‘clearinghouse’ to be established
The order also states that secretaries and directors of several agencies will work together to form an AI cybersecurity ‘clearinghouse’ “in voluntary collaboration with the AI industry and operators of critical infrastructure”.
The role of the clearinghouse will be to coordinate scanning for software vulnerabilities and to distribute vulnerability patches when they are discovered.
The Office of Management and Budget, Department of Homeland Security and other agencies are also expected to release “binding operational directives” and other guidance to federal government agencies under the order.
These will cover “expediting and prioritising the cyber defence of civilian federal government information systems”; establishing or expanding federal programmes and cybersecurity services “that enhance AI-enabled defensive tools”; and enabling federal government agencies, state and local authorities, and operators of critical infrastructure to access cybersecurity tools and services, including frontier AI models where appropriate.
Other measures outlined in the order include agencies working to determine whether any federal grant programmes have funding available that can be directed toward applicants developing advanced AI vulnerability detection; and expanding the “hiring and placement pathways” of the federal government’s Tech Force information cybersecurity specialists.
About Jack Aldane
Jack is a British journalist, cartoonist and podcaster. He graduated from Heythrop College London in 2009 with a BA in philosophy, before living and working in China for three years as a freelance reporter. After training in financial journalism at City University from 2013 to 2014, Jack worked at Bloomberg and Thomson Reuters before moving into editing magazines on global trade and development finance. Shortly after editing opinion writing for UnHerd, he joined the independent think tank ResPublica, where he led a media campaign to change the health and safety requirements around asbestos in UK public buildings. As host and producer of The Booking Club podcast – a conversation series featuring prominent authors and commentators at their favourite restaurants – Jack continues to engage today’s most distinguished thinkers on the biggest problems pertaining to ideology and power in the 21st century. He joined Global Government Forum as its Senior Staff Writer and Community Co-ordinator in 2021.
