Anthropic is minimizing the concerns that prompted the White House to prohibit the export of its advanced “myth” and “fable” AI models. U.S. officials suggest this is indicative of CEO Dario Amodei’s self-interested and unilateral approach to cybersecurity.
Recently, Amazon CEO Andy Jassy reportedly alerted the Trump administration about researchers “jailbreaking” Fable, raising potential national security alarms.
As a result, the White House imposed export controls on Mythos and Fable, preventing their use outside the U.S. In contrast, Anthropic opted to take both models offline entirely.
Even though Anthropic claims the vulnerability is a “narrow escape possibility” not exclusive to Fable, the company disagrees with the White House’s handling of the situation.
These conflicting statements drew criticism from Trump administration officials, who noted that Anthropic had been fanning the flames of fear regarding AI disasters for months. One senior official remarked, “They seem more focused on their safety than anyone else’s. If I were a human investor, I’d be quite worried.”
Another official disclosed that the Trump administration had heard from Amazon and nearly six other companies expressing their worries about Fable before deciding to act. However, the identity of the companies wasn’t revealed.
The second U.S. official suggested that if Anthropic had genuinely addressed the problem, perhaps by temporarily suspending access instead of treating it as a minor issue, things could have turned out differently.
Top leaders in the administration felt they could not trust Anthropic to resolve these concerns. This distrust intensified after a call between Amodei and senior Trump officials. Reports indicate that company executives declined requests to collaborate with government experts.
Just last week, Amodei wrote a blog post stating that the U.S. government “should have the power to block” any AI models considered overly dangerous.
Officials close to Anthropic claimed that the Trump administration had contacted them on Friday, giving them just 90 minutes to take down the model.
The official argued against reports that Amodei was hard to reach, insisting that both he and other senior executives were available within 15 minutes during White House hours.
“There was no refusal to address the issue. We didn’t get any specific details,” the official said.
Anthropic is keen to reinstate its models and is reportedly working diligently to meet the government’s demands.
In a blog post, Anthropic stated it had little choice but to comply with export regulations, since some employees are foreign nationals who would not be allowed to work on the models.
Treasury Secretary Scott Bessent and White House Cyber Secretary Sean Cairncross were involved in heated conversations with Antrhopic’s leadership on this matter.
David Sachs, the White House AI adviser, expressed confusion regarding Anthropic’s reluctance to adhere to previously established safety measures, calling it “frankly perplexing.”
He added that the hope is for Anthropic to rectify the safety issues and for the export limits to be lifted, enabling Fable’s return to the market.
The decision by the White House was partly influenced by worries that groups associated with China might have secretly accessed Mythos. Anthropic has already restricted access to its products in China.
Initially, Anthropic provided Mythos to about 40 leading firms including Apple and JPMorgan in a pilot program, cautioning that wide access could lead to severe cyberattacks. Nearly two months later, they introduced a “secure” version of Mythos, called Fable 5, to the public.
Perry Metzger, president of an AI policy group in Washington, D.C., commented, “If you were human, what outcome did you expect? You’ve been warning people about the dangers of your job while asking the government to intervene.”
Since April, Anthropic has asserted that its “Claude Mythos” AI model is too perilous for public release, arguing for the necessity to pause development for safety’s sake.
Amodei is known for making alarming predictions about AI risks, once stating there’s a 25% chance it could lead to humanity’s downfall. He has also warned that the national unemployment rate might drastically rise due to AI technologies.
Ben Narasin, of Tenacity Venture Capital, mentioned that Anthropic’s alarming claims seem to have backfired.
“He’s become all too predictable by always crying wolf. Now you got what you wanted, but did you realize it’s not what you actually wanted?” Narasin said, though he acknowledged that Anthropic could still find ways to benefit from the chaos.
He pointed out the challenges facing competitors like Dalio and OpenAI, suggesting that any resulting constraints might simply reinforce Anthropic’s position in the industry.
This warning led to accusations from Sachs and others that Anthropic employs fear tactics to influence regulations in ways that favor the company over its competitors.
Yet, doubts linger as the company is in the midst of striving for growth. Recently, Anthropic quietly filed for an IPO ahead of its bigger rival, OpenAI, aiming to leverage the public market for cash while collaborating with major companies like SpaceX and Google within its data centers.
