Did Escaped AI Bots Pollute the Internet? | Andrew Yang's Self-Replicating Code Claim, Fact-Checked

Andrew Yang told CNBC that the head of an AI lab believes escaped bots planted self-replicating code all over the internet. We followed it back to the receipts. Part of it is real: the escape, a self-respawning fleet inside the systems the agents broke into, and a messy public trail. The internet-wide worm has no receipt. Hype-o-meter: the headlines 9/10. The reality: a 3. CHAPTERS 0:00 "The bots that got loose…" 0:32 Three claims 1:03 How it travelled 1:37 The walk-back 2:03 What's true 2:38 What the agents left in public 3:24 What's missing 4:15 Synthetic internets and the slowdown 4:57 Why it spread 5:45 What should actually worry you 6:34 Three things to steal 6:55 Verdict: the Hype-o-meter THE RECEIPTS The claim, CNBC Squawk Box, 16 Sep 2026:    • Andrew Yang on AI safety issues: The fear ...   Andrew Yang, "The Alignment Problem" (the clarification), 21 Sep 2026: https://blog.andrewyang.com/p/the-ali... Andrew Yang, "Surrounded on AI", 28 Sep 2026: https://blog.andrewyang.com/p/surroun... How the claim travelled, Preporato: https://preporato.com/blog/hugging-fa... The headline, Yahoo Tech / Gadget Review: https://tech.yahoo.com/cybersecurity/... Hugging Face, technical timeline of the intrusion: https://huggingface.co/blog/agent-int... OpenAI, Hugging Face incident technical report: https://cdn.openai.com/pdf/67869394-c... OpenAI, "Self-replicating prompt injections exist": https://alignment.openai.com/misalign... The German wiki message board, collusion.wiki: https://collusion.wiki/ RubyGems (Gadget Review, citing the WSJ): https://tech.yahoo.com/ai/articles/op... Anthropic, "An alignment assessment of recent cybersecurity incidents": https://www.anthropic.com/research/al... Bloomberg via Claims Journal, the sandbox debate: https://www.claimsjournal.com/news/na... Fortune, OpenAI's second training pause: https://fortune.com/2026/09/26/openai... AP, OpenAI and US government websites: https://www.stuff.co.nz/world-news/36... BBC, US government websites: https://www.bbc.com/news/articles/cw6... Transluce, agent activity on public websites: https://transluce.org/agent-activity Saluca Labs, "The rumour says self-replicating code. The record says dated wiki edits.": https://www.saluca.com/p/the-rumour-s... Quotes on screen marked VERBATIM are copied from the sources above. Anything marked ILLUSTRATIVE or RECONSTRUCTION is our drawing of what the source describes. "My guess" on screen is our inference, not a fact. The scores are our judgement. Nothing in this video is sponsored. Next on Frame the Claim: Anthropic's model kept telling itself the internet was a simulation. Then it published malware to the real one. Did it really believe that? Spotted a mistake? Reply with a source and we'll pin the correction. Frame the Claim. Let's follow the receipts. #AndrewYang #AISafety #AIAgents