Per Woofun AI, Nicholas Carlini, a leading security expert at Anthropic, has pivoted from advocating for a release delay to actively lobbying the White House following extensive testing of the Mythos model. Carlini, formerly of Google and known for embedding hidden commands in classical music to control Amazon Alexa, observed Mythos identify 479 Linux vulnerabilities and automatically generate exploit code within days, outperforming human experts. He developed the 'Carlini Loop' prompting technique to ensure varied results, while the model also uncovered 500 vulnerabilities in Ghost web publishing software in two weeks.

This rapid discovery rate triggered widespread panic in the security community, dubbed 'Bugmageddon,' exacerbated when an official Ghost patch was reverse-engineered by hackers to breach over 700 websites by April 2026. The incident underscored the AI security paradox: vulnerabilities are found in seconds, but human patching takes weeks. Following a jailbreak vulnerability in Fable 5 flagged by Amazon's security team and a call from CEO Andy Jassy, the White House issued an emergency shutdown order against Anthropic last Friday. Carlini was subsequently dispatched to Washington to demonstrate security measures and argue that deploying the defense version of the model is safer than containment.