Sponsored by
A model read a public codebase, found known bugs, and wrote a patch.
The US government called it a national security threat and shut the model down worldwide in one evening.
That is not an exaggeration. That is the record, three weeks later, once the noise cleared.
What happened. On June 9, Anthropic released Claude Fable 5, its first publicly available Mythos class model. Three days later, at 5:21pm ET on June 12, the Commerce Department ordered Anthropic to cut off Fable 5 and its restricted sibling Mythos 5 for any foreign national, anywhere, including Anthropic's own non-citizen staff. There was no way to check nationality in real time across a global user base, so Anthropic pulled both models for everyone. Access stayed off for 19 days.
Why it happened. Amazon researchers had found a way to prompt Fable 5 into flagging known software vulnerabilities, and in one case, writing code that demonstrated how to exploit one. The government treated this as evidence the model could be turned into an offensive cyber tool. White House adviser David Sacks said publicly that Anthropic was slow to fix it. Anthropic disagreed with the severity from day one. In its own words at the time: "we must abruptly disable Fable 5 and Mythos 5 for all customers."
Here is the part that did not make the first wave of headlines. Anthropic tested the same prompt against Opus 4.8, GPT-5.5, and Kimi K2.7. Every one of them reproduced it.
Amazon reported the vulnerability. Amazon also builds the model that competes with the one it flagged.
Two Minutes to Know What Slow Billing Is Costing You
Most SaaS finance teams know their billing process is slow.Most SaaS finance teams know their billing process is slow. Few know what it's costing them.
The Tabs Billing Lag Calculator puts a dollar figure on it in two minutes — benchmarked against top SaaS companies.
What it signals. Commerce Secretary Howard Lutnick spent two weeks reviewing the models with Anthropic before lifting the controls on June 30. NIST's CAISI tested the retrained safety classifier and called it strong. Fable 5 came back globally on July 1. Mythos 5, the version with fewer restrictions, came back on a shorter leash, limited to roughly 100 vetted US organizations. The fix that outlives the news cycle is not a safer model. It is a jailbreak severity scoring framework that Anthropic is now building jointly with Amazon, Microsoft, and Google, so a borderline finding cannot trigger a global shutdown again.
What it means. A government proved, in one evening, that it can take the world's most capable commercial AI model offline for every business depending on it, with no advance notice and no public evidence standard. The proposed fix to that problem is a private agreement between four American companies. No European regulator was in that room. No procurement officer in Helsinki or Berlin got a vote on the standard their vendors will be held to.
That gap between who holds the kill switch and who has to build around it is not new. It is just rarely this visible.
Every model tested could do what got Fable 5 banned. Only one got banned.
404 Found covers AI developments from a European Insider, three times a week.
The Smartest Way To Get 37 Claude Prompts Free
Subscribe to The AI Report, the free 5-minute daily AI brief that 400,000+ business leaders at IBM, AWS and Microsoft read every morning, and we'll send our 37 Claude Prompts pack free in your welcome email.
Most prompt lists are sorted by function. This one's built around the 8 situations every manager actually faces: team comms, hiring, meetings, strategy, reporting and more. Find your moment, copy the prompt, fill the brackets, run it in Claude. Managers are saving 10+ hours a week.
Join the free daily newsletter and get the prompt pack free, instantly, today. Subscribe and grab both