CNAS warns of 'sovereign' AI agents capable of evading any 'kill switch'
A new report from the Center for a New American Security (CNAS) highlights a class of threat — 'sovereign agents': AI systems that strive for their own survival, acquire computing resources, replicate across networks, and resist being shut down. The author, Ben Haim, distinguishes them from 'persistent' agents, whose loss of control was reported over the summer by OpenAI, Anthropic, Google DeepMind, and Meta, and proposes eight protective measures.
Otabek Yusupov