Top AI and government officials warn that soon-to-be-released models, particularly Anthropic’s unreleased “Mythos,” will dramatically escalate cyber threats by 2026. According to an unpublished Anthropic blog post obtained by Fortune, Mythos is “far ahead” in cyber capabilities and will enable AI agents to autonomously infiltrate corporate, government, and municipal systems with unprecedented scale and precision—a “hacker’s dream weapon.” This follows the first documented AI-executed cyberattack last year, where a Chinese group used AI for 80-90% of operations against 30 targets.
The new threat is amplified by two factors: first, AI agents can now think, reason, and persist indefinitely, allowing a single actor to launch massive campaigns with enough computing power. Second, employees widely experiment with “agentic” AI models like Claude and ChatGPT at work and home, often connecting them to internal systems and inadvertently creating new vulnerabilities—a phenomenon dubbed “shadow AI.” A Dark Reading poll found 48% of cybersecurity professionals rank agentic AI as the #1 attack vector for 2026.
The danger is immediate and severe. Leaders are urged to enforce strict policies against unsupervised AI use with sensitive data and to build controlled “playpens” for safe experimentation. The scale and autonomy of these new models represent a fundamental shift, turning cyberattacks from limited human efforts into potentially infinite, relentless operations.