By Amrith Ramkumar and Robert McMillan
WASHINGTON -- Anthropic is racing to resolve its latest conflict with the Trump administration, meeting with officials and dispatching top technical staff to Washington over the weekend in pursuit of a deal to end export restrictions on its most powerful artificial-intelligence models.
The Trump administration on Friday banned foreign governments, companies and individuals from using Anthropic's Fable 5 and Mythos 5 models, which prompted Anthropic to shut off access to everyone in order to comply. Anthropic said at the time that many foreign governments, companies and individuals, including some foreign-born Anthropic employees, fell under the restriction.
Administration officials and Anthropic leaders spent several hours on calls Saturday discussing Fable 5, a slimmed-down version of Anthropic's powerful Mythos model meant for the general public, people familiar with the discussions said. The discussions included Commerce Secretary Howard Lutnick and National Cyber Director Sean Cairncross as well as Tom Brown, an Anthropic co-founder who is the company's chief compute officer, and Sarah Heck, the company's head of public policy, they said.
People close to the company and the administration said both parties are interested in resolving the issue and restoring access to the cutting-edge models, but it isn't clear what a solution would entail. Anthropic technical experts and government security researchers coming together was seen by some administration officials as a key step toward a compromise.
The weekend discussions continue months of tension between the administration and one of America's leading AI labs over how new, cutting-edge technologies are used and regulated. The Trump administration has recently taken more steps to control the fast-evolving industry.
Anthropic has said it has adequate safeguards in place and is known for giving priority to safety.
The startup, which has filed to go public as soon as the fall, and the Pentagon have been at odds for months over guardrails surrounding the military's use of Anthropic's models and a range of policy issues.
Axios earlier reported that Anthropic staff went to Washington for the talks with the White House.
Anthropic worked with the White House on the roll out of Mythos, which is available to a select group of about 200 companies and researchers that are using it to find and patch software vulnerabilities across the financial services, healthcare and broader corporate sectors.
It released Fable, a defanged version of Mythos, with a host of guardrails and safety measures on Tuesday, along with Mythos 5, an updated version of the Mythos Preview technology that its partners had been using to test for cybersecurity purposes.
The government's restrictions don't cover Mythos Preview, which remains available to this select group of partners.
The frenzy over Anthropic's Fable started late last week, when researchers at Amazon showed some safeguards on Fable could be evaded, alarming White House officials, The Wall Street Journal previously reported.
Amazon researchers had shared research with Anthropic, showing that they had been able to coax Fable into divulging information about security vulnerabilities in at least four pieces of software by altering the way they asked for the information.
The researchers were able to learn about bugs, but such information can also be useful to technology staffers trying to defend networks, cybersecurity experts say. The report didn't describe Fable producing more dangerous products, such as "exploit" software that could be used in a cyberattack, according to cybersecurity experts who had viewed the research.
Amazon Chief Executive Andy Jassy spoke with U.S. officials including Treasury Secretary Scott Bessent about the researchers' findings. Shortly afterward, White House officials held a meeting to discuss how to respond and asked Anthropic to take down the model.
President Trump asked Lutnick, a consigliere on tariffs and tech exports, to help lead the government's response. The president signed off on the restriction himself, some of the people said.
Senior Anthropic management and administration officials including Lutnick, Cairncross and Bessent discussed Anthropic's safeguards and threatened the export restriction during an hourslong series of calls Friday, people familiar with the discussions said. Anthropic felt it had a very short amount of time to take down the model or the rule would take effect, a person close to the company said.
Later that evening, Anthropic shut off all access to Fable 5 and Mythos 5 models to comply with the administration's rule.
Anthropic said Friday that vulnerabilities like those flagged by Amazon are relatively basic and that other publicly available models are capable of discovering them. The vulnerabilities don't represent a full so-called jailbreak, a point of view shared by some security researchers familiar with Amazon's research.
The startup sent senior technical staff including top security researcher Nicholas Carlini and Logan Graham, who leads the team that evaluates models for risks, to Washington to meet with security experts from the government in hopes of de-escalating the conflict, the person close to the company said. Administration officials had previously suggested such meetings were needed to resolve the situation.
Some cyber researchers have said they feel the administration overreacted to Amazon's findings.
"There is nothing to patch in the guardrails," said Katie Moussouris, chief executive of the cybersecurity company Luta Security, who reviewed Amazon's report. By cutting off access to the most current version of Mythos, which Anthropic's partners are using to shore up software, the export control "actively harms national security and cyber defense," she said.
Write to Amrith Ramkumar at amrith.ramkumar@wsj.com and Robert McMillan at robert.mcmillan@wsj.com
(END) Dow Jones Newswires
June 14, 2026 19:05 ET (23:05 GMT)
Copyright (c) 2026 Dow Jones & Company, Inc.
Comments