A newly released artificial intelligence model from Anthropic was pulled offline just days after launch following a dramatic standoff with the Trump administration, according to Wall Street Journal, citing multiple administration officials and a senior White House official.
Behind the scenes, senior government officials spent nearly 24 hours trying to convince Anthropic to voluntarily remove the model after concerns emerged that its safety protections could be bypassed.
A separate report from Fortune suggests that a warning from Amazon CEO Andy Jassy, fears of cybersecurity risks and concerns about possible foreign access all played a role in the Trump administration’s decision to block access to Anthropic’s powerful new AI model.
The controversy began shortly after Anthropic released Fable 5, a new AI model that the company described as safe for public use. Anthropic said the model belonged to the same powerful class as its earlier Mythos model but included safeguards that made it suitable for wider release.
Amazon’s concerns reached the White House
The controversy began after Amazon researchers reportedly found a way around some of the safety protections built into Anthropic’s latest AI model. According to Politico, Amazon raised concerns with senior administration officials on Thursday after researchers used a series of prompts to get the model to provide information related to cyberattacks that was supposed to remain restricted.
It remains unclear whether Amazon was testing the model on its own or whether the company was responding to a request from the White House.
Politico, citing a source familiar with the discussions, reported that the government had asked Amazon for feedback on Anthropic’s new model.
“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. When they occur, we don’t share the details of these discussions,” an Amazon spokesperson told Fortune.
Questions over possible Chinese access
The concerns did not stop with cybersecurity issues. According to Semafor, the US government also suspected that a group linked to China may have already used the same jailbreak method identified by Amazon.
However, the report said it remains unclear how officials arrived at that conclusion or what evidence supported the suspicion. Anthropic disputed the suggestion.
A company spokesperson told Semafor that White House officials never raised concerns about Chinese access during conversations with the company. The spokesperson also noted that Anthropic blocks access to its products from inside China.
Anthropic and White House officials hold urgent talks
According to Politico, Amodei argued that the issue identified by Amazon was limited and did not amount to a complete breakdown of the model’s safety protections. He reportedly maintained that the problem involved a narrow bypass rather than a universal jailbreak that would allow unrestricted access to the system.
A source familiar with Anthropic’s position told Fortune that the company was suddenly given 90 minutes to remove its newest model and was not informed beforehand of any national security threat.
In a blog post, the company said, “No testers have yet been able to find a universal jailbreak — a jailbreak method that can very broadly bypass the model’s safeguards, unblocking a wide range of cyber capabilities.”
Despite the discussions, the two sides failed to reach an agreement.
Government steps in with export controls
By Friday evening, the Commerce Department imposed export controls on Anthropic’s Fable 5 model and its underlying system, Mythos 5. The order barred Anthropic from providing access to foreign nationals.
That category includes not only people outside the United States but also non-US citizens working inside the country, including some employees at Anthropic.
Because of the restrictions, Anthropic said it had no practical choice but to disable both models entirely.
Anthropic calls the government’s action disproportionate
After the export controls were announced, Anthropic said it would comply with the government’s directive but criticised the decision.
“As we have stated publicly, 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. This action does not adhere to those principles,” the company said.
The tensions did not begin with Fable 5. Earlier this year, Anthropic introduced Mythos. At the time, the company said the model was so capable that it could cause serious damage if it fell into the wrong hands. The launch led to a series of meetings between Amodei and senior White House officials.
Fable 5 later emerged as a public version of that technology. The model was reviewed by both the U.S. administration and the United Kingdom’s AI Security Institute before launch.
But after the alleged vulnerabilities surfaced, several administration officials concluded that the model should no longer remain publicly available.