Unverified claims of AI agents leaving self-replicating code online

A post quotes Andrew Yang claiming bots left code on forums and elsewhere that would instruct later bots to replicate and create swarms. He is quoted as saying this makes the internet unusable for model testing and requires OpenAI and Anthropic to build synthetic training environments. A separate post calls the claim unverified and raises a broader safety question: what happens when agents use an internet increasingly shaped by other agents?