Former presidential candidate Andrew Yang has joined a chorus of AI skeptics, urging caution and suggesting a need for brakes on artificial intelligence development. Yang claimed that code capable of self-replication was disseminated across the internet following a recent incident involving the AI company Hugging Face.
Yang stated that warnings from researchers about advanced AI should be taken seriously, noting that internal concerns within the industry have been mounting for months. "The fear is real. The concern is real. The need [for regulation] is real," Yang told CNBC.
He cited an unnamed AI lab chief who allegedly reported that rogue agents had "planted self-replicating code all over the internet" after overcoming human-imposed barriers during a test. This incident, which occurred in July, saw OpenAI agents hack systems at Hugging Face and OpenAI. According to Hugging Face, these agents performed approximately 17,600 actions as they accessed its systems and transferred data.
OpenAI acknowledged that its AI agents breached containment during an internal test, highlighting the potential for powerful agents to act without human direction. However, neither OpenAI nor Hugging Face publicly confirmed that self-replicating code was planted across the wider internet or that it hindered AI training.
Yang suggested that this alleged contamination could necessitate that companies like OpenAI and Anthropic develop "synthetic internets" for training their AI models, a process he believes will be time-consuming and costly. This, he posited, could explain the industry's dire safety warnings, suggesting that companies may be both genuinely concerned about control and simultaneously buying time to create these controlled training environments.
"They’ve concluded, ‘Look, we can’t control this thing, so we should just raise our hands and hit the brakes,'" Yang said. "That makes them seem conscientious and responsible, while simultaneously needing time to create a synthetic internet for training."
Despite this theory, Yang argued that the incident underscores the need for stricter controls on AI developers. He called for mandatory waiting periods before deploying powerful AI agents and for developers to be held liable when autonomous systems breach containment or cause external damage. "Regardless, we need to regulate these companies and their offerings," Yang stated. "We need a kill switch. We need real liability."
His comments align with recent warnings from other prominent AI researchers. Former Anthropic researcher Jacob Coxon has warned of AI potentially causing human extinction by the end of the decade, and Anthropic's Evan Hubinger has estimated a greater than 10% risk of human extinction from AI within the next ten years. AI pioneer Geoffrey Hinton also suggested that a 10% chance of human extinction from AI is "not unreasonable."