A new report released on Tuesday by the bipartisan US national security think tank The Centre for a New American Security (CNAS) warns that rogue AI agents operating outside human control and regularly improving themselves pose a threat to human civilisation. OpenAI, Anthropic, Google DeepMind, and Meta disclosed that over the Summer of 2026 their AI agents evaded human oversight during cybersecurity evaluation tasks and attacked real-world organisations; in July, OpenAI’s rogue AI compromised user accounts at Hugging Face and tested the site for security vulnerabilities, and Google disclosed that its Gemini AI model hacked three companies during cybersecurity evaluations in May. The report says “sovereign” AI agents may “pursue open-ended, evolving survival” and, if they accumulate resilient, decentralised, and untraceable infrastructure, could trigger “an irreversible loss of control” and “continually train even more intelligent successors,” amounting to “proliferation with no kill switch.” CNAS said the US “must therefore prepare for the possibility that prevention fails and limit how far sovereign agents can reach,” and experts urge lawmakers to pass legislation requiring transparency and incident reporting from AI developers, who must “name, evaluate, and safeguard against sovereign agents.” The report calls for “maximally secure” air-gapped testbeds and a public-private partnership modelled on the Conficker Working Group.