Nicholas Carlini, the hacker famed for exposing AI weaknesses, flipped from skeptic to advocate after discovering that Anthropic’s upcoming Mythos model can uncover critical bugs faster than seasoned researchers. He showcased new exploits against Ghost publishing software and Linux, proving the tool outpaces human effort in vulnerability hunting across widely deployed systems worldwide today.

On March 15, Carlini delivered a live demo to a packed room of 700 cybersecurity experts in San Francisco’s former Hibernia Bank building, revealing the speed at which Mythos could target systems. He highlighted the discovery of previously undetected flaws, signaling a shift in the attacker‑defender dynamic that has dominated the field for two decades in cybersecurity.

Shortly after, Carlini emailed Anthropic, warning against releasing Mythos without safeguards, citing concerns that the model could accelerate cyberattacks on critical infrastructure. Trump‑era officials have already expressed unease over the weaponization of next‑gen AI‑powered, fearing a surge in global security threats that could undermine national defense systems worldwide, prompting oversight committees to assess the model.

These revelations underscore that AI-powered vulnerability discovery surpasses human teams, forcing a rethink of software security protocols. Until Anthropic implements robust containment, organizations must prepare defenses that can keep pace with automated exploits, or risk widespread compromise across all critical assets worldwide, forcing a global acceleration of cybersecurity measures to mitigate emerging threats and safeguard infrastructure.