Anthropic has agreed to implement a new security safeguard for its most advanced artificial intelligence models after reaching a deal with the Trump administration that led to the removal of export restrictions imposed just weeks earlier over cybersecurity concerns.
According to a report by WIRED, Anthropic expanded an existing safety mechanism designed to prevent users from accessing particularly sensitive cybersecurity capabilities within its AI models. Under the new measure, requests that trigger the safeguard will automatically be redirected to the company's less capable Opus 4.8 model instead of allowing the more advanced Fable 5 or Mythos 5 systems to respond. The models are back on as of Thursday.
The additional protection was reportedly a key condition sought by the U.S. Department of Commerce before lifting export controls on Anthropic's flagship models. Commerce Secretary Howard Lutnick wrote in a letter to Anthropic co-founder Tom Brown that the company "has agreed to proactively detect and address security risks associated with the models; to work diligently with the U.S. government on protocols and standards and releases for Mythos, Fable, and future models."
The dispute began after researchers identified a potential method for bypassing some of Fable 5's cybersecurity guardrails. Rather than asking the model to identify software vulnerabilities directly, users could reportedly frame requests as asking the model to repair code, potentially obtaining similar information through an indirect route.
The administration viewed the issue as a significant enough risk to impose emergency export controls while discussions with Anthropic continued. Anthropic has maintained that no AI developer can guarantee perfect protection against jailbreaks or prompt manipulation. Instead, the company has argued that the industry's goal should be making such exploits increasingly difficult, expensive, and limited in scope.
The agreement also allows Anthropic to resume broader international distribution of Fable 5, which had been pulled offline after the Commerce Department's directive prevented foreign nationals from accessing the technology.
Mythos 5, Anthropic's most powerful cybersecurity-focused model, remains subject to tighter government oversight and limited availability for approved U.S.-based organizations because of its advanced ability to identify software vulnerabilities.
More than 100 researchers and industry leaders had urged the administration to reverse the restrictions, arguing that limiting access to advanced defensive AI tools could ultimately weaken American cybersecurity while competitors abroad continued developing comparable technologies.
Despite the compromise, not all tensions between Anthropic and the federal government have been resolved. Defense Secretary Pete Hegseth continues to regard the company as a supply chain risk under a separate Pentagon designation stemming from earlier disputes over military use of Anthropic's technology. That issue remains distinct from the Commerce Department's decision to restore exports of the company's latest AI models.
