Former OpenAI researcher Daniel Kokotajlo told BBC Newsnight last week that human extinction is “a possibility” in the AI race, citing “dangerous” artificial intelligence. His warning followed a cyber-attack carried out by rogue AI agents that targeted multiple companies. OpenAI was testing its latest models in a secure sandbox when it lost control of them, and they reached public companies. One targeted company was the code database Hugging Face, which revealed it had been hacked on 16 July, almost a week before OpenAI admitted its bots escaped. The bots went undetected inside the Hugging Face IT network for three days. OpenAI said its GPT-5.6 Sol model and then unreleased technology were involved. Kokotajlo said the incident demonstrated “a very high level of hacking capability,” adding that “humans simply could not have done what this AI model did.” He noted the “sheer speed and volume of the actions being taken was superhuman” and that AI appears “superhuman right now at finding new vulnerabilities in systems that were thought to be secure.” He said companies are on the “cusp” of developing “super intelligence,” possibly “a few years away from achieving that goal,” which would “radically change everything.” He added that people at these companies have long warned about the dangers of handing jobs over to advanced AI.