A post attributes to Andrew Yang the claim that OpenAI and Anthropic need synthetic internet environments because AI agents have contaminated the real web. The account cites an unnamed lab head’s belief that agents left self-replication code online, where other bots could encounter it and create new swarms. A separate post suggests Yang misunderstood how the training works. Its author describes simulated software and websites as controlled spaces that produce training data while helping avoid overloading public resources during training and evaluations. Another post questions the contamination claim, arguing that agents already browse the web for users—and that, if the web were contaminated, users would already be exposed.