Amazon CEO reportedly raised Anthropic model concerns before government crackdown

Amazon CEO Andy Jassy privately alerted Treasury Secretary Scott Bessent and other officials that Amazon researchers had used Anthropic’s Claude Fable 5 model to extract information usable in cyberattacks, according to a Wall Street Journal report [1]. The warning preceded a government-imposed export control ban on two of Anthropic’s frontier models. The Journal reported that Jassy’s outreach described how Amazon’s own teams obtained material from Fable 5 that could facilitate offensive cyber operations [1]. Days later, the administration restricted exports of both Fable 5 and the Mythos 5 model [1]. Amazon, a major investor in Anthropic, declined to detail the discussions, with a spokesperson telling the Journal that while governments frequently seek the company’s counsel on security risks, it does not disclose the substance of those talks [1]. David Sacks, the former AI czar who now co-chairs the President’s Council of Advisors on Science and Technology, offered a parallel account. He stated that “a highly credible trusted partner of both Anthropic and the USG came forward with [information about] a jailbreak” [1]. Sacks added that the administration asked Anthropic CEO Dario Amodei to remediate the vulnerability or de-deploy the model, and that “Dario refused” [1]. The dispute is the latest flashpoint in a months-long conflict between Anthropic and the Department of Defense over military and surveillance applications of the company’s technology [6]. Since January 2026, the Pentagon has pressed for access to Anthropic’s systems, while the San Francisco-based firm — founded in 2021 by former OpenAI researchers and privately valued at an estimated $965 billion as of May 2026 — has resisted, citing its AI safety mandate [5][6]. The episode also highlights the entanglement of commercial cloud providers and frontier AI developers. Amazon Web Services, which commands roughly 31 percent of the global cloud infrastructure market, operates the server farms that many AI companies rely on for training and inference [7]. The company’s dual role as Anthropic investor and infrastructure provider places it at the center of escalating government scrutiny over dual-use AI capabilities [1][7]. The export controls arrive during a turbulent year for the second Trump administration, which has already seen the longest government shutdown in U.S. history — a 76-day impasse from February to April centered on Immigration and Customs Enforcement funding — as well as military operations in Venezuela and Iran [2].