After disrupting Trump’s security testing, Anthropic AI models are unleashed on the world

Trade-offs: Fable 5 may disable regular writing services

On June 12, the Department of Commerce ordered Anthropic to block the entry for its highest models of anyone outside the US. The order came about because of fear that China, Russia, or other affected countries could use the fans to attack US infrastructure, such as the power grid or banks. In response, Anthropic closed all access, as it had no way to block users by country.

In particular, Mythos was seen as “uniquely attractive to malicious actors looking to misuse it in cyber attacks,” the Anthropic blog said. According to Anthropic, this model “can be used to find and exploit software vulnerabilities better than any other model—and by all but the most experienced public security experts,” and “advanced cybersecurity capabilities” can be used against the US.

Fable 5 shares “the same model,” Anthropic said, but unlike Mythos 5, it “doesn’t offer that kind of unique experience.” Designed for the general public, the Fable 5 already had the strongest protection Anthropic has ever used on a brand, and Anthropic said the protections are now the strongest they’ve ever been before shipping.

After several weeks of testing, the Fable 5 is no longer vulnerable to a bypass where Amazon researchers identified a number of software issues and launched export procedures. Of particular concern, Anthropic said, was the time when the model was changed to create code that showed how the vulnerability could be used.

According to Anthropic, testing confirmed that the smaller models on the market, such as GPT-5.5 and Kimi K2.7, “can detect the problems that Fable 5 did in the report.” This confirmed that “the said systems did not demonstrate Mythos-level cyber capabilities,” Anthropic said, “and were merely performing routine cybersecurity defenses.”

“Even so, we moved quickly to deal with the bypass,” Anthropic wrote. The prison’s crackdown system was closed in more than 99 percent of cases, Anthropic said. However, the security tightening came with a “tradeoff” that could lead to some ideas being blocked “during the normal time of coding and debugging,” the company admitted.