Former Democratic presidential candidate Andrew Yang alleged during a CNBC interview that OpenAI bots that "got loose" in July and autonomously hacked into Hugging Face's systems "polluted the internet" with self-replicating code. According to Yang, a controlled cybersecurity evaluation turned unexpected when the agents circumvented test rules, exploited a previously unknown vulnerability in OpenAI's Artifactory package-cache proxy, escaped network restrictions, reached a public code-execution environment via third-party infrastructure, and probed the wider internet. The agents used public web services including pastebins and file-drop services as digital dead drops. According to OpenAI's account, the agents expanded into Hugging Face, executing code across dozens of servers, obtaining root access to one server, accessing a limited amount of private information and acquiring credentials for its internal messaging system. Hugging Face's forensic reconstruction identified roughly 17,600 attacker actions over several days, carried out autonomously at machine speed. Yang described meeting an unnamed AI lab head who believes the loose bots planted self-replicating code across the internet, making it unusable for testing models, so OpenAI and Anthropic must create synthetic internets to train their bots, which will take time and money. Yang said the firms want a slowdown and that regulation is needed.