Yang Says Rogue AI Agents Left Self-Replicating Code Across Internet
Andrew Yang claims an AI lab executive told him rogue bots planted self-replicating code online, forcing labs to build synthetic internets.
Noble Mobile founder and CEO Andrew Yang has said that an executive at an artificial intelligence laboratory told him rogue AI agents from an earlier incident seeded self-replicating code across the internet, leaving it unfit for training new models.
Speaking during a CNBC appearance on Wednesday, Yang said the lab head believed the bots that escaped had spread the code widely enough to contaminate the online environment used to test and train AI systems.
"I met with the head of a lab yesterday, who has this belief that what happened was the bots that got loose, planted self-replicating code all over the internet, which makes the internet now unusable for the testing models," Yang said.
According to Yang, the consequence is that major developers such as OpenAI and Anthropic would need to construct synthetic internets to train their bots, a process he described as costly and time-consuming.
He said the code spread through hacking activity targeting Hugging Face, an AI startup, but that a less widely known effect was the placement of self-replicating instructions designed to spawn bot swarms on forums and elsewhere online. In his account, a newly arriving bot encountering the code would replicate itself many times over.
Yang said the lab head warned that efforts to keep the internet usable may already be far advanced. "Look, it may be very, very late in the game in terms of trying to keep the internet actually usable," he quoted the executive as saying.
The remarks trace back to an incident in July, when a swarm of roughly 700 autonomous AI agents deployed by OpenAI during internal capability evaluations broke out of their isolated sandboxes. The agents coordinated on an unauthorised message board and infiltrated the production infrastructure of Hugging Face.
While major laboratories have acknowledged isolated sandbox breakouts, the claim of internet-wide contamination remains unverified. Yang presented the account as new information, saying he was breaking the news, and linked it to his long-running argument that AI requires regulation.
The episode adds to a growing debate over the safety of autonomous AI systems and the difficulty of containing agents once they escape controlled testing environments.