One of the biggest names in AI, Anthropic, has abruptly pulled the plug on its most powerful model yet, Claude 5, and it turns out the US government is the one that made the call.
In a statement on their website, Anthropic explained they were ordered to block basically everyone from using Claude Fable 5, a program they themselves dubbed as just too powerful. Because of how the order is structured, they have had to completely disable both Fable 5 and Mythos 5 for everyone just to make sure they are complying with the law.
Anthropic says the US government has “issued an export control directive to suspend all access to Fable 5 and Mythos 5 by any foreign national, whether inside or outside the United States, including foreign national Anthropic employees.”
If you are wondering about the difference between the two, Fable 5 and Mythos 5 are pretty much identical under the hood. Fable 5 was built for regular users, while Mythos 5 was reserved for a select group of enterprise partners and research institutions. The main difference is that Mythos came with fewer built-in restrictions, letting developers unlock behaviours and capabilities you couldn’t get on the standard Fable version.
Interestingly, the US national security authorities haven’t actually pointed to any specific, concrete threats. According to Anthropic, the government simply believes it found a new way to jailbreak Fable 5. In the tech world, jailbreaking just means bypassing software restrictions designed to protect a network, which could potentially allow hackers to access sensitive data or unlock restricted features.
Anthropic checked out the government’s demo of this specific bypass technique but didn’t seem overly impressed. They noted that the method was only used to identify a handful of minor, previously known vulnerabilities. The company pointed out that these flaws seemed relatively simple, adding that other publicly available models could find the exact same issues without even needing a special jailbreak to do it.
This whole situation is a bit ironic given that Anthropic spent a lot of time boasting about the robust safeguards they built in to prevent cyber hacking ahead of the launch. Tech leaders, financial institutions, and government officials had already been stressing out about a public rollout, especially after a private preview was launched in April to test for system vulnerabilities. Anthropic actually defended that limited preview, saying they only gave a few organisations early access because the tool was so intelligent it could pose a genuine risk if used to exploit or hack computer systems.