Amazon Ratted Out Anthropic to the White House
Friday night’s government shutdown of Fable 5 and Mythos 5 already smelled political. Now we know who lit the match.
The Wall Street Journal reports that Amazon CEO Andy Jassy personally took cybersecurity research to the White House — research that claimed Fable 5 could be prompted into revealing information useful for cyberattacks. Shortly after that conversation, the Commerce Department issued the export control directive that killed both models.
Let me translate that: Amazon develops its own AI. Amazon also sells cloud services to Anthropic. Amazon’s CEO went to the White House about a rival’s product. Two days later, that product was dead.
Timing is a coincidence the size of a mountain.
The “jailbreak” doesn’t hold up to scrutiny. Anthropic says the government showed them a “narrow, non-universal” technique: get Fable 5 to review a specific codebase for bugs. Security researcher Katie Moussouris — founder of LutaSecurity — looked at the Amazon paper and posted on Bluesky: “I’ve seen the paper. It’s not a jailbreak.” Even Anthropic points out that GPT-5.5 finds the same class of vulnerabilities without any bypass.
If every model that could review code for bugs was shut down, we’d have no models. That’s the entire value proposition of an AI that can read code — which Amazon’s own CodeWhisperer and Bedrock services also do.
But here’s where it gets ugly. Former Commerce Department official Kate Koren told the WSJ something worth sitting with: the White House’s existing dislike of Anthropic may have influenced the decision. And let’s not pretend that’s baseless. Anthropic refused to let its AI power mass surveillance or lethal autonomous weapons. The administration retaliated in February by banning federal agencies from using Anthropic’s AI and labeling the company a “supply chain risk.” Anthropic sued.
So the timeline shapes up like this: company refuses government’s AI demands → government punishes company → Amazon hands government a pretext → government uses it.
The strongest counterargument: National security is national security. If there’s even a chance Fable 5 accelerates offensive cyber capabilities for adversaries, you don’t wait for peer review. You act. The White House’s Axios-sourced comments say the real need is time to “harden” the national security apparatus — a process they claim could finish in weeks, not months. If that’s true, this is a temporary pause with a clear endpoint, not a permanent kill. That changes the calculus.
I’ll give that argument its due. If the shutdown is genuinely about hardening and lifts in weeks with a transparent process, I’ll say I was wrong to be so cynical. But the burden of proof is on the administration to show that — and right now, they’re not trying.
Also worth saying: Amazon’s security research might have been legitimate. Maybe their researchers genuinely found something concerning and escalated it through proper channels. The problem isn’t that Amazon flagged concerns. The problem is that the government treated a contested and narrow finding as a kill switch — with no due process, no adversarial testing, no chance for Anthropic to respond before the axe fell.
And the irony? Semafor reports the White House also imposed controls on Mythos because of suspicions a China-linked group had accessed it. So we have Amazon research + China fears + a pre-existing grudge = the first AI model execution by government fiat.
We’re going to see a lot more of this. Every AI company with a cloud provider competitor should be taking notes. If you’re building frontier models on AWS, your landlord is also your rival — and they have the White House on speed dial.
The Friday letter was bad enough. This makes it worse.
Sources: The Verge — Amazon security research reportedly led to the White House’s Anthropic Fable ban (Terrence O’Brien, Jun 14, 2026), Ars Technica — Anthropic shuts down Fable, Mythos models following Trump admin directive (Kyle Orland, Jun 13, 2026)