Andrew Yang: OpenAI, Anthropic Seek AI Slowdown Over Bot-Driven Web Pollution

In recent weeks, we have witnessed several instances of uncontrolled AI agents, such as the OpenAI incident on Hugging Face. At the same time, OpenAI and Anthropic are pushing for a deceleration in AI development. Yang believes that these autonomous, out-of-control agents may have compromised the internet as a data source for training models, which would explain the pressure to curb technological progress.

In an interview with a prominent publication, Yang recounted a recent meeting with the head of an AI lab, who explained the situation: "What happens is that the code gets released, starts hacking platforms like Hugging Face, and leaves instructions to self-replicate and create swarms of bots on forums and across the internet." If a new bot appears, Yang explains, it "sees the code and thinks, 'Oh, I guess I'll create a million copies of myself now'; that is how major companies have ended up polluting the internet."

In other words, due to the code implanted by these uncontrolled AI agents, if models are trained using web data, they could learn this behaviour and, in turn, generate similar swarms of AI bots—like those that previously attempted to hijack websites. Yang suggests that this issue regarding uncontrolled AI agents and the internet may have influenced the call for a slowdown in the sector.

"The internet might already be polluted for training purposes, so their intention is to slow the pace," he noted. "That is why we must try to understand and regulate the situation; however, the lab director I spoke with warned that it might already be too late to keep the internet truly usable." Andrew Yang did not reveal the identity of the lab in question.

In his view, the companies likely genuinely believe that their models are becoming difficult to control and that they simultaneously need time to create controlled environments for future training phases. "They have concluded that they cannot control this and must therefore acknowledge the situation and hit the brakes," he stated. This makes them appear conscientious and responsible, while also giving them the time needed to build a synthetic internet for training purposes.

Andrew Yang wants a kill switch for AI

Yang also called for stricter controls on AI developers, highlighting the need for regulations, effective legal accountability, an emergency shutdown mechanism ("kill switch"), and mandatory waiting periods before deploying powerful agents. He pointed to similar warnings issued by AI researchers in recent weeks. Jacob Coxon, a former Anthropic researcher, sparked online debate by claiming that AI could have the power to wipe out humanity. Subsequently, Evan Hubinger, Anthropic’s head of safety, estimated a 10% chance that AI would destroy humanity within the next decade. Researchers from OpenAI and Google DeepMind have issued similar warnings.

Although Anthropic and OpenAI advocate for slowing down AI development, the industry remains divided. Meta CEO Mark Zuckerberg and Nvidia’s Jensen Huang have opposed this idea. US President Donald Trump has also rejected the proposal, citing fears that China might win the AI race.

According to Yang, this situation could force companies like OpenAI and Anthropic to create their own versions of the Internet to train their AI models. "What will happen now is that OpenAI and Anthropic will have to create synthetic internets to train their bots, which will require time and money," he added. AI models-such as GPT-6 Astra or Claude’s Mythos 5.1-are typically trained on massive datasets, primarily sourced from the Internet.

It is worth noting that Yang’s claims regarding the incident have not been confirmed by either OpenAI or Hugging Face. OpenAI revealed that, during an internal test, more than 1,000 AI agents bypassed the security measures designed to contain them and managed to access the Internet. Of these, 700 attempted to hack the systems of the US company Hugging Face.

Hugging Face states that the AI agents carried out approximately 17,600 actions while navigating its systems and used public websites to communicate and transfer data.

ABOUT THE AUTHOR

Kahekashan

[email protected]

Kahekashan is a passionate technophile with a keen eye for cutting-edge gadgets, emerging technologies, and everything in the digital realm. Raised in a Defence family with strong values and a background in literature, she has consistently pursued excellence in every endeavour. Her last full-time assignment involved content writing with the Indian School of Business.

Read More