Andrew Yang says rogue AI bots have polluted the Internet

Andrew Yang says rogue AI bots have polluted the Internet

Former US presidential candidate Andrew Yang says that rogue AI agents may have made the internet useless to train AI. This, he says, may be why OpenAI and Anthropic want to slow down AI development.Former US presidential candidate Andrew Yang has a theory over calls for AI slow down from OpenAI and Anthropic. (Photo: Reuters)In recent weeks, we have seen several cases of rogue AI agents, such as OpenAI’s Hugging Face incident. At the same time, OpenAI and Anthropic are pushing for a slowdown in AI development. Former US presidential candidate Andrew Yang believes rogue AI agents may have turned the internet useless for AI companies to train their models, leading to this push for a slowdown.In an interview with CNBC, Yang said he recently met the head of an AI lab who told him about this situation. “What happened is the code gets loose, it goes around hacking Hugging Face, the (rogue AI agents) left code to self-replicate and create bot swarms on forums and around the internet,” If a new bot shows up, Yang says, it “sees the code they're like oh like I guess I'm gonna now create a million of myself and so now the major firms have polluted the internet.”That is, because of the code planted by rogue AI agents, if models are trained online, they may learn this behaviour, potentially creating similar swarms of AI bots that have previously tried to hijack websites. Andrew Yang says that this situation around rogue AI agents and the internet may have played a role in the push for an AI slowdown.“The internet now may be polluted for training purposes so what they want to do is slow down,” he said. “So this is why we have to try and get our arms around and regulate but the head of the lab that I spoke to said look it may be very very late in the game in terms of trying to keep the internet actually usable.” Andrew Yang did not name who this head of AI lab was. In his view, the companies may genuinely believe their models are becoming harder to control, while also needing time to build controlled environments for further training. "They've concluded, 'Look, we can't control this thing, so we should just raise our hands and hit the brakes,'" he said. "That makes them seem conscientious and responsible, while simultaneously needing time to create a synthetic internet for training."Andrew Yang wants a kill switch for AIYang also called for tighter controls on AI developers, saying there should be regulation, real liability, a "kill switch" and mandatory waiting periods before powerful agents are deployed. He pointed to similar warnings from AI researchers in recent weeks. Former Anthropic researcher Jacob Coxon sparked discussions online, stating that AI may have the power to end humanity. Anthropic safety lead Evan Hubinger later predicted that there was a 10 per cent chance that AI may end humanity within the next decade. Similar warnings have been given by researchers from OpenAI and Google DeepMind. While Anthropic and OpenAI are pushing for an AI slow down, the industry remains divided. Meta CEO Mark Zuckerberg and Nvidia’s Jensen Huang have pushed back. US President Donald Trump has also rejected such a proposal, citing concerns that China may win the AI race.This situation, Yang says, may force companies like OpenAI and Anthropic to create their own forms of internet to train AI models. "So what happens now is that OpenAI and Anthropic have to create synthetic internets to train their bots, which is gonna take some time and money,” he added. AI models like GPT-6 Astra or Claude’s Mythos 5.1 are usually trained on large datasets, primarily via the internet.Keep in mind that Yang’s claims have not been confirmed by OpenAI or Hugging Face regarding the incident. OpenAI has disclosed that during an internal test, over 1,000 AI agents got around safeguards meant to contain them and gained access to the internet. 700 of those agents tried to hack systems of the US company Hugging Face.Hugging Face says the AI agents carried out about 17,600 actions as they moved through its systems and used public websites to communicate and transfer data.- EndsPublished By: Armaan AgarwalPublished On: Sep 18, 2026 12:32 IST

Original Source

Read the full article at Indiatoday →

KhanList aggregates and links to publicly available news content. We do not host full articles from third-party sources. Always verify important information with original sources.