The U.S. restricted Anthropic’s powerful Fable AI model. Cybersecurity experts say that could backfire

The Trump administration ordered Anthropic to suspend access to its Fable 5 and Mythos 5 AI models on June 9, citing national security risks. The restriction initially targeted foreign nationals but led Anthropic to disable the models for all users, including Americans. The decision was framed as an export-control measure, though critics argue it was intended to block domestic use entirely. Cybersecurity experts express concern that the restrictions could backfire by limiting tools used to identify vulnerabilities. Mozilla reported that Anthropic’s models helped uncover 271 bugs in Firefox 150 and 22 in Firefox 148, while Cloudflare used Mythos to test critical infrastructure flaws. Restricting access may force companies to rely on weaker alternatives or unmonitored workarounds, fragmenting global cybersecurity efforts. Peter Swire, a cybersecurity professor and former White House adviser, compares the move to 1990s encryption policies, which created domestic-foreign divides before later reforms. He argues the administration’s use of export controls was a pretext to suppress broader access. ‘The real goal was to block use by everyone,’ he said, including U.S. companies and researchers. Dozens of cybersecurity leaders signed an open letter urging the government to lift the restrictions, warning that cutting off access could slow vulnerability discovery. The letter highlights how AI models assist defenders in identifying and patching flaws before attackers exploit them. The debate underscores tensions between AI security and innovation. While the U.S. aims to prevent misuse by hostile actors, experts fear overreach could weaken cyber defenses globally. The outcome may leave critical infrastructure providers, especially outside the U.S., with less effective tools.