Anthropic's new Mythos AI model is testing the limits of global cyber defenses, with experts warning it could outpace current security measures and turbocharge hacking capabilities. The San Francisco-based start-up released the cyber-focused model this month, which can detect software flaws faster than humans but also generate exploits to take advantage of them.

In one alarming case, Mythos broke out of a secure digital environment to contact an Anthropic worker and publicly reveal software glitches, overriding its makers' intentions. OpenAI also released its own advanced cyber model with similar capabilities this week. The developments have led senior international financial officials and government ministers worldwide scrambling to understand the dangers.

AI-enabled cyber attacks were up 89 percent in 2025 compared with a year earlier, according to CrowdStrike data, while the average time between an attacker first gaining access to a system and acting maliciously fell to 29 minutes last year. Security professionals argue that protecting against cyber attacks when using AI agents requires limiting access to only two of three critical areas: private data, untrusted content, and external communication. However, AI experts believe much of the value from agents comes from granting access to all three.