On June 30, the Commerce Department lifted the export controls that had forced Anthropic to switch off Claude Fable 5 and Mythos 5. Today, Fable 5 returns. The models went dark for eighteen days, and the reason they went dark deserves your attention.
On June 12, at 5:21 p.m. EDT, Anthropic received a private letter citing national security authorities. It barred any foreign national, inside or outside the United States – including Anthropic’s own non-citizen employees – from accessing the models. With no way to verify nationality in real-time, the company pulled Fable 5 and Mythos 5 for everyone. A commercial product deployed (in Anthropic’s words) to hundreds of millions of people went offline.
The stated concern was a jailbreak. Amazon researchers had prompted Fable 5 into flagging software vulnerabilities and, in one case, generating exploit code. Anthropic tested the claim and found that Claude Opus 4.6, Opus 4.8, OpenAI’s GPT-5.5, and China’s Kimi K2.7 could all do the same thing. The capability that triggered a national security action already ships in half-dozen models you could use the whole time, which raises the question: What was the actual reason behind the shut down?
If you believe that we are in an AI arms race, the eighteen-day blackout carried a real cost. While the U.S. throttled its most capable model, Chinese open-source developers kept closing in on the American frontier at a fraction of the price. Washington handed them three weeks of clear road. (Remember, we are living on the exponential. Three weeks is an insane amount of exponential improvement time.)
Here’s the problem: Anthropic brought Fable 5 back with the same reasoning core and a tighter leash. The redeploy added a cybersecurity safety classifier that Anthropic says blocks the exact vulnerability-hunting technique from the Amazon report in more than 99% of cases. Sounds good so far, but… the classifier is aggressive by design, and Anthropic admits it now flags benign requests more often during routine coding and debugging. When it trips, you get a notification and your request is answered by Claude Opus 4.8, which Anthropic itself groups among its less capable models, so on security-adjacent coding and debugging work, the frontier model you came for will more often hand you a prior-generation answer. Fable 5’s reasoning is unchanged, but the system now catches a lot of legit dev traffic and routes it to Opus 4.8.
The federal government showed it can reach into the market and turn off a frontier AI product with a letter, no public process, and no published standard, then reverse itself once the facts caught up (if they actually did).
Not for nothing, Anthropic is completely cleared to use the most powerful AI in the world to create AI-native versions of anything they decide to focus on (business, research lab, product, etc.). Think people without access to the best models can compete? Think again.
Every company needs a Claw strategy. Do you have one?
Author’s note: This is not a sponsored post. I am the author of this article and it expresses my own opinions. I am not, nor is my company, receiving compensation for it. This work was created with the assistance of various generative AI models.