Anthropic Security Expert Reverses Stance, Becomes White House Lobbyist After Mythos Model Discovers 479 Linux Flaws

According to Beating, Nicholas Carlini, a top security expert at Anthropic, shifted his position dramatically after testing the company's new Mythos model. Previously known for criticizing AI safety delays, Carlini discovered that Mythos identified 479 Linux kernel vulnerabilities in just days—flaws he had never found himself—and generated attack code automatically. Concerned about the risks, he initially sent a warning memo urging the company to delay release.

The vulnerability discoveries triggered widespread alarm in the security community, dubbed "Bugmageddon." By April 2026, over 700 websites were breached by attackers exploiting newly discovered flaws. After Amazon's CEO Andy Jassy warned the White House of jailbreak vulnerabilities in a competing model, the administration issued an emergency ban on Anthropic. However, Carlini was subsequently deployed to Washington as a company lobbyist, now arguing to government officials that releasing a secured version is safer than keeping the model locked away.