Amazon CEO Andy Jassy reportedly warned the Trump administration about security risks in Anthropic's Fable model, joined by executives from other tech companies.
Amazon appears to have triggered the chain reaction against Anthropic's most advanced AI model. The Information and Axios report that Amazon CEO Andy Jassy, along with other tech executives, brought concerns about the security risks of Anthropic's Fable model directly to senior Trump administration officials.
On Thursday evening, Amazon handed the government a report claiming to show how parts of the Fable model can be unlocked through jailbreaking. What makes this remarkable: Amazon is one of Anthropic's biggest investors. Anthropic trains and runs its models on Amazon's own AI chips, among other infrastructure. A major investor effectively turning in its own portfolio company to the government is, to put it mildly, unusual.
"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," an Amazon spokesperson told Axios.
According to Axios, at least five other companies besides Amazon reached out to senior government officials on Thursday evening and Friday morning. National Cyber Director Sean Cairncross responded by calling a meeting with top White House officials.
For hours, the government tried to convince Anthropic to pull the model voluntarily. Anthropic refused. The White House didn't send the official export control order until 5:20 p.m. ET, giving the company just 90 minutes to comply. By 10:00 p.m., Anthropic had shut the model down.
Cybersecurity expert Katie Moussouris, who reviewed the Amazon report at Anthropic's request, called the government's response wildly disproportionate. What Amazon flagged was just "Defense Oriented Prompting" (DOP), she said, a technique used by defenders rather than attackers, not a jailbreak. "If national security is the goal, this is an own goal against us," she wrote on LinkedIn.
A source told Axios the escalation may have had less to do with the actual security risk than with the broader relationship between Anthropic and the U.S. government. Officials were bothered by what they saw as a "lack of seriousness" in how Anthropic handled the Fable release. The export restriction only came after Anthropic ignored an earlier request to pull the model on its own.
"Companies will not screw with the White House. That is the ultimate effect," a person familiar with the situation told Axios, calling it a "de facto licensing regime." According to The Information, the export restriction is "unlikely" to be extended to other AI labs.
Shortly before, the administration had issued an executive order calling for a security review that now appears less voluntary than advertised. Anthropic publicly welcomed the order but went ahead and released Fable before the review process was even in place, a move that likely added to the administration's frustration with the company.
Some observers see the escalation as a direct result of Anthropic's own aggressive marketing around the Mythos model's alleged—and likely real—cybersecurity capabilities. If you keep telling the world how dangerous your own technology is, don't be surprised when the government takes you at your word. Even the name fits. A fable is a short story with a moral lesson.
That reading isn't wrong, but it doesn't go far enough. What happened here is a massive government intervention in the private sector, aimed at bringing one of the most powerful AI technologies under state control while cutting other countries off from access. Anthropic and the U.S. government have clashed publicly several times recently, and the possibility of politically motivated pressure on the company can't be dismissed.
Subscribe to THE DECODER for ad-free reading, a weekly AI newsletter, our exclusive "AI Radar" frontier report six times a year, full archive access, and access to our comment section.
此内容由惯性聚合(RSS阅读器)自动聚合整理，仅供阅读参考。 原文来自 — 版权归原作者所有。