Anthropic has regained permission to distribute its flagship artificial intelligence models internationally after agreeing to implement additional cybersecurity safeguards demanded by the Trump administration, marking one of the first major agreements between Washington and a leading AI developer over export controls tied to advanced generative models.
The agreement, first reported by WIRED, ends a weeks-long dispute that began after the U.S. Department of Commerce imposed emergency export restrictions on some of Anthropic's most advanced AI systems following concerns that their cybersecurity capabilities could be exploited through prompt manipulation.
Under the revised arrangement, Anthropic has expanded an existing security mechanism governing its latest AI models. Instead of allowing sensitive cybersecurity requests to reach its most capable systems, qualifying prompts will now be automatically redirected to a less powerful model designed to provide safer responses. Following the agreement, the company's Fable 5 models resumed broader availability on Thursday.
The compromise illustrates the increasingly active role the U.S. government is taking in regulating frontier artificial intelligence. Rather than permanently restricting advanced AI exports, federal officials negotiated technical safeguards intended to reduce the likelihood that sophisticated cybersecurity capabilities could be misused while allowing American companies to remain competitive in global markets.
According to WIRED, the Commerce Department viewed the new safeguard as a critical requirement before removing the export restrictions. Commerce Secretary Howard Lutnick outlined the government's expectations in a letter to Anthropic co-founder Tom Brown.
In the letter, Lutnick wrote 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 originated after researchers demonstrated a possible method for circumventing some of Fable 5's cybersecurity protections. Rather than directly requesting software vulnerabilities, users could reportedly frame questions as requests to repair or improve code, potentially obtaining similar technical information through an indirect approach.
Federal officials concluded that the vulnerability warranted temporary export restrictions while negotiations continued. Anthropic has acknowledged that preventing every possible jailbreak or prompt injection remains technically impossible, arguing instead that developers should focus on making such attacks increasingly difficult, expensive and limited in effectiveness.
The revised safeguard reflects that philosophy. Requests deemed especially sensitive will now be routed away from Anthropic's most capable frontier models-including Fable 5 and Mythos 5-and instead handled by the less advanced Opus 4.8 system, reducing the likelihood that users can access the company's highest-end cybersecurity capabilities.
While the agreement restores international access to Fable 5, Anthropic's most powerful cybersecurity model remains subject to tighter controls. Mythos 5 will continue operating under more restrictive government oversight and remain available only to approved U.S.-based organizations because of its advanced ability to identify software vulnerabilities.
The restrictions had prompted concern across the artificial intelligence industry. More than 100 researchers and technology leaders reportedly urged the administration to reconsider the export controls, arguing that limiting access to American-developed defensive AI tools could inadvertently weaken U.S. cybersecurity leadership while rival companies overseas continued advancing comparable technologies.
The Commerce Department ultimately chose a negotiated solution rather than maintaining broad restrictions, a move that could establish a framework for future interactions between regulators and AI developers as increasingly capable models emerge.
Despite the agreement, Anthropic's relationship with the federal government remains complicated. The Commerce Department's decision applies specifically to export controls governing advanced AI models, but it does not resolve separate concerns raised by the Defense Department.
According to the report, Defense Secretary Pete Hegseth continues to classify Anthropic as a supply-chain risk under an unrelated Pentagon designation connected to previous disputes involving military applications of the company's technology. Those issues remain under separate review and were not affected by the Commerce Department's export decision.
