Amazon just quietly triggered one of the biggest AI policy fights in Silicon Valley. According to the Wall Street Journal, the company's cybersecurity researchers discovered that Anthropic's advanced AI model Fable 5 could be manipulated into revealing information useful for cyberattacks. Amazon CEO Andy Jassy then brought those findings directly to White House officials. Within weeks, the government issued an export control directive that forced Anthropic to cut off access to Fable 5 and Mythos 5 — effectively locking out foreign nationals from using technology their own company created.
Here's where it gets messy. Anthropic isn't some tiny startup — it's one of the most respected AI safety companies in the world, founded by former OpenAI executives. The company has consistently pushed back against government pressure to weaponize AI, refusing to let its models be used for mass surveillance or autonomous weapons. But that principled stance appears to have made it a target. Just this year, Trump ordered federal agencies to stop using Anthropic's AI entirely, and his Defense Secretary Pete Hegseth flagged the company as a "supply chain risk." Now, with Amazon's research in hand, the White House had the justification it needed to tighten the screws further.
The security research itself is contested. Anthropic disputed the government's framing, arguing that calling Fable 5's vulnerabilities a "jailbreak" is misleading — and that similar risks exist in other publicly available models like GPT-5. Even some respected security researchers agree. Katie Moussouris, founder of LutaSecurity and a prominent voice in cybersecurity policy, posted on BlueSky that after reviewing Amazon's paper, "It's not a jailbreak." A former Commerce Department official speculated to the WSJ that the White House's existing animosity toward Anthropic may have colored how they interpreted Amazon's findings. Translation: maybe this was less about genuine security concerns and more about finding a pretext to restrict a company that won't play ball.
The collateral damage is real. Many of Anthropic's own researchers are foreign-born, meaning they're now barred from accessing their own product. Imagine building something at work and then being told you can't use it because of where you were born. It's a stark reminder that AI policy isn't just about technology — it's about which companies get government favor and which ones don't. Amazon, meanwhile, sits in a uniquely powerful position. The company has massive AWS cloud infrastructure, deep ties to government, and now has demonstrated it can weaponize security research to influence policy against competitors.
What happens next will signal whether this is about genuine national security or political punishment. If the White House uses the same standard to restrict other AI companies that have similar vulnerabilities, then maybe there's a legitimate policy at work. But if Anthropic remains singled out while rivals like OpenAI face no such restrictions, it becomes clear this is really about control. Either way, the message to AI companies is unmistakable: cooperate with government demands, or face export bans and supply chain designations. For an industry that claims to care about safety, that's a chilling precedent.