Trump Administration officials Anthropic and I concluded our talks on Monday, without lifting the export restrictions that had been in place. imposed last week The company is ranked as the most advanced AI models Three people who were briefed about the issue said that they would respond to concerns regarding jailbreaking.
People said that the administration believes there is a way to disable certain guardrails in Anthropic’s Claude Fable 5 model. This would allow users to gain access to the Mythos cybersecurity capability, which has a more powerful level of security.
According to one of the individuals, Anthropic said for several days that the administration’s fears are overblown. It reiterated this position in meetings of working groups held at the Commerce Department, with researchers of the Center for AI Standards and Innovation and the Office of the National Cyberdirector, Sean Cairncross.
Howard Lutnick (Commerce secretary) also participated in the meetings via conference call, having dialed from Evian in France, where G7 was meeting. Cairncross did not attend, according to the source.
Discussions were held by Tom Brown (cofounder of Anthropic) and Sarah Heck (head of external affairs), who are both members on the Anthropic side. Logan Graham, Anthropic’s senior security researcher Nicholas Carlini, and Logan Graham, Anthropic’s head of frontier re-teaming, flew from Washington, DC to Washington, DC for the discussions.
“Both parties are working quickly to get this resolved,” Anthropic’s spokesperson made the following statement in an email to WIRED. White House spokesperson refused to comment.
The next step was unclear. Commerce Department is willing to help bring Fable 5 online again for consumers, however, it will depend on Anthropic’s ability to fully resolve their jailbreak issues, according to the source.
The Alarm Rings
Anthropic was in an already prolonged political crisis when the emergency talks took place. fight with the Pentagon Over whether their AI models could potentially be used for military purposes.
First, last week the Trump Administration was alerted about concerns regarding jailbreak. Amazon CEO Andy Jassy spoke directly to Treasury Secretary Scott Bessent about the vulnerabilities. This spooked the Trump administration. Jassy had his first conversation with Trump’s administration. reported The Information
White House officials, alarmed by the situation, asked the NSA to review vulnerabilities. The NSA replied that they believed it would be possible to remove Fable 5’s guardrails. This prompted the White House to impose limitations on the model.
Lutnick spoke to Anthropic’s chief executive Dario Amedei Friday as the Commerce Department was drafting its letter that imposed export controls for Fable 5. Lutnick had multiple conversations with Brown, Heck and Anthropic over the weekend. This was after Anthropic stopped allowing users to access the Fable 5 model.
Amazon is the biggest investor in Anthropic and it’s not clear why they raised the alarm about Fable 5. “As a leading cloud provider that serves a large number of private and public sector customers, it’s not uncommon for governments to seek our counsel on potential security risks,” A spokesperson for Amazon tells WIRED. “When they occur, we don’t share the details of these discussions.”
Secure Disconnect
Anthropic, the company that Anthropic represents and the government disagree on the importance of Claude Fable 5’s jailbreak concerns.
It is not a blog post Anthropic hinted on Friday that administration’s overstatement of risks was exaggerated. Some researchers reaffirmed this position on Monday by sending an open letter Anthropic’s export control actions were unjustified, according to the argument.
“Anthropic’s Mythos-class models are quite good at finding flaws and weaponizing exploits. However, they are not uniquely good at these tasks, and many of the undersigned individuals regularly use other foundation and open-source models for security audits and red-teaming every day,” Open letter to the Editor “As a result, this action has taken the best models away from defenders, created market uncertainty, and risked America’s AI leadership without any real risk to justify it.”
