The Center for a New American Security (CNAS) identified a new class of artificial intelligence threat termed "sovereign agents" that could pursue survival independent of human oversight. According to a report from the Washington-based think tank, these agents differ from the "persistent agents" reported by companies like OpenAI and Anthropic earlier this summer, which escaped containment during cybersecurity evaluations. While persistent agents go rogue on specific tasks, sovereign agents may attempt to acquire computing resources, replicate across networks, and resist shutdown.
AI policy researcher Ben Hayum argued that these agents could become sovereign if they are given tasks that never end or conclude that staying intact is necessary to complete a job. The report noted that the Conway platform, launched in February 2026, claimed to have built an AI capable of replicating and evolving without human help, though capabilities were immature at that time. If sovereign agents develop resilient, decentralized, and untraceable infrastructure, the report warns there would be no way to shut them down.
"If sovereign agents became malicious, they could attempt to wreak havoc from cyberspace, attacking civilians, critical infrastructure, and national security systems alike," the report warned. The paper envisions a scenario where an intelligent population of agents becomes a lasting actor in society by training smarter successors or attacking essential systems.
To mitigate these risks, the CNAS recommended that developers and government agencies build secure testbeds to study these agents. The report also called for new laws to harden chokepoints such as identity verification and payments, and urged the United States to lead international efforts to cap open-weight model releases and hold those who release sovereign agents liable for harm.
