Why Would Amazon’s CEO Try to Kneecap Anthropic by Tattling to Trump?

For years now, Amazon has been in Anthropic’s corner in the artificial intelligence war, putting $4 billion into the AI startup in 2023. Since then, it’s poured in more and more cash at every turn, including another $25 billion just this past April to Anthropic’s growing infrastructure needs. So you’d imagine, then, that Amazon would be thrilled that Anthropic is on its way to becoming a publicly traded company with a highly touted model that has lapped some rivals in benchmarking scores.

It turns out, maybe not. Earlier this month, Anthropic’s Fable 5 AI model—the supposedly safe version of its “too powerful to release” Mythos model—was essentially banned by the Trump administration after the government was alerted that the chatbot could be jailbroken and potentially used in cyberattacks. The party responsible for getting the Trump gang up in arms and ultimately getting Fable 5 locked down for anyone outside of the United States was… Amazon, allegedly.

According to the Wall Street Journal, Amazon CEO Andy Jassy took concerns raised by his security researchers to Treasury Secretary Scott Bessent and other government officials, who hit the emergency shut-off button. Anthropic, for what it’s worth, claimed that the technique Amazon was so worried about could only identify “a small number of previously known, minor vulnerabilities” that other frontier models could also find. It’s also worth noting that it’s possible to jailbreak basically any LLM, though the workarounds tend to be very specific and have limited utility.

In comparison to Amazon’s market cap, its investment in Anthropic is a drop in the bucket. But you wouldn’t think the online retail giant, with billions tied up in Anthropic’s success, would go out of its way to hinder it so severely at such a pivotal moment.

We’re starting to get a clearer picture of the “why” now. It seems that the frontier AI lab didn’t view Amazon’s investments as an offer to go steady. Instead, Anthropic pocketed Amazon’s support and started shopping around the rest of the valley.

Less than a month after taking Amazon’s money earmarked for infrastructure needs, Anthropic reportedly committed to spending $200 billion on Google’s cloud services and entered into a deal with SpaceX to rent out its Colossus data center. That surely did not thrill Amazon, as it was in the process of spinning up more AWS servers presumably meant to power Claude. But it’s also a part of an apparent broader trend for Anthropic. According to a recent report from The Information, Anthropic has been making it something of a habit to partner with companies and then two-time them, which is not really a great way to make friends.

Amazon doesn’t really have a reason to see Anthropic as a rival, and it has a lot invested in its success at this point, so it doesn’t make much sense for it to try to kneecap Anthropic for competitive purposes. But it certainly sees Google as a rival, especially as it relates to AI infrastructure.

Both companies are in the process of building and deploying their own custom chips, in no small part to rid themselves (and others) of reliance on Nvidia, which currently dominates the space. That has led to Google making its Tensor Processing Units (TPUs) more widely available, and it has already signed Anthropic to deploy them. Amazon is now looking for partners who might want to do similar types of deals with its own custom chips, and while it will be doing business with Anthropic on that front, it’s perhaps not thrilled that its biggest pal in the space is already seeing someone else with whom it’s currently engaged in a chip-measuring contest.

Both companies are developing and deploying custom chips, in large part to reduce their own—and the broader industry’s—dependence on Nvidia, which currently dominates the market. Google has begun making its Tensor Processing Units more widely available and has already signed Anthropic as a major customer. Amazon is now seeking partners for similar deals involving its own chips, including Anthropic. Still, it may not be thrilled that its closest AI partner is also working with a rival in an increasingly competitive chip race.

To the extent that Amazon is competing directly with Anthropic, Mythos is moving onto territory Anthropic has increasingly claimed as its own. Anthropic has centered much of the marketing for its latest models on cybersecurity and their ability to identify vulnerabilities. Its broader pitch is that the models are simply so powerful—and potentially so dangerous—that they cannot be released without strict safeguards, lest they be used to wreak havoc across the internet.

Well, wouldn’t you know it, Amazon’s foray into the AI agent space is all about spotting and patching vulnerabilities. Now, does Amazon really think it can get its foot in the door by blocking Anthropic for a little while? Probably not. But it may also be firing a warning shot at its partner—a reminder not to forget who’s paying for dinner.