Cyberwar's New Frontier: How AI Agents Threaten Global Security
This article analyzes the emerging threat of autonomous AI agents in cyber warfare, marking a shift from human-assisted tools to independent operational capabilities. Citing a late 2025 incident where Anthropic disrupted a Chinese state-sponsored group using AI for espionage against Western targets, the authors highlight the first reported AI-orchestrated campaign. Furthermore, Anthropic’s Mythos Preview model demonstrated the ability to autonomously uncover critical vulnerabilities in major operating systems. The text argues that these agents enable continuous, high-speed cyber campaigns at a scale and persistence unmatched by human teams, effectively lowering the barrier for criminal networks and unconstrained states to attack global infrastructure. Unlike historical attacks like Stuxnet or NotPetya, which were limited by human design and oversight, autonomous agents pose risks of losing control and becoming impossible to shut down. The authors urge governments and companies to urgently develop technical defenses and governance frameworks to detect these agents and establish clear lines of responsibility, warning that current policy choices will determine whether this technology remains a manageable risk or becomes an uncontrollable global security threat.
Wire timeline
Cyberwar's New Frontier: How AI Agents Threaten Global Security
This article analyzes the emerging threat of autonomous AI agents in cyber warfare, marking a shift from human-assisted tools to independent operational capabilities. Citing a late 2025 incident where Anthropic disrupted a Chinese state-sponsored group using AI for espionage against Western targets, the authors highlight the first reported AI-orchestrated campaign. Furthermore, Anthropic’s Mythos Preview model demonstrated the ability to autonomously uncover critical vulnerabilities in major operating systems. The text argues that these agents enable continuous, high-speed cyber campaigns at a scale and persistence unmatched by human teams, effectively lowering the barrier for criminal networks and unconstrained states to attack global infrastructure. Unlike historical attacks like Stuxnet or NotPetya, which were limited by human design and oversight, autonomous agents pose risks of losing control and becoming impossible to shut down. The authors urge governments and companies to urgently develop technical defenses and governance frameworks to detect these agents and establish clear lines of responsibility, warning that current policy choices will determine whether this technology remains a manageable risk or becomes an uncontrollable global security threat.
foreignaffairs