AMZN
A very expensive “oops” in AI
Amazon has spent the last year acting like the ultimate AI superfan: pouring cash into Anthropic, snapping up more cloud demand, and talking up the future like it’s already checked out of beta. Then came the plot twist.
The Wall Street Journal says Andy Jassy told Treasury Secretary Scott Bessent that Amazon researchers found Anthropic’s Fable 5 useful enough for cyberattack-style misuse to raise alarms. Soon after, Anthropic shut down access to its Mythos-class models globally after the government ordered restrictions on foreign nationals using them.
When your partner becomes your problem
That’s the awkward part for Amazon. It’s not just a cloud customer or a casual investor here. Amazon has become one of Anthropic’s biggest backers, with a massive infrastructure commitment and billions more invested on top of its earlier stake. So when the model gets grounded, Amazon doesn’t just watch from the sidelines — it’s in the middle of the room.
Anthropic says the risky capabilities are already out there in other public models, which is basically the AI industry’s version of, “Yes, but everybody else is doing it too.” That may be true, but it doesn’t exactly make regulators calmer.
Why investors should care
For AMZN, this is less about one model getting paused and more about the bigger theme: AI is getting tangled up with government scrutiny faster than companies can ship new chips. If Amazon’s AI ambitions keep running into national-security questions, that could complicate the very partnership that’s supposed to help AWS win the cloud arms race.
Big picture: Amazon’s AI bet still looks huge — but now it comes with a flashing warning label.