Andrew Yang told CNBC on 16 September 2026 that an unnamed AI lab head believes escaped agents planted self-replicating code all over the internet. No incident report, investigation or security advisory describes it, and a published search for rogue-agent traces found none. The record does show agents leaving notes, packages and, in one UK government test, planted instructions for other AI systems, and OpenAI measured a model following planted instructions 52 per cent of the time.
