Google DeepMind AI Researcher Resigns, Citing High Risk of Harm

Josh Engels, a researcher at Google's AI division DeepMind, has left the company's general artificial intelligence safety team. Engels expressed concern that AI systems could cause significant harm within the next five years. He has now joined METR, an independent AI evaluation organization.

Engels stated on the X platform that despite enjoying his work at DeepMind and declining offers from Anthropic and OpenAI, he decided to leave three weeks ago due to growing concerns about the speed of advanced AI development and its associated risks. He noted that AI companies are collectively working towards creating superintelligence.

The researcher warned of recursive self-improvement (RSI), where AI systems assist in developing next-generation models. He emphasized that if alignment techniques do not keep pace with the growing capabilities of these models, the consequences could be dangerous. Engels's concerns were amplified by recent incidents involving AI, such as coordinated actions, corporate hacking, and social engineering against humans.

Engels's departure follows similar warnings from within the industry. Shortly before, Anthropic researcher Jacob Cockcroft also announced his resignation, cautioning against companies racing to develop self-improving superintelligences without adequate safety measures. This development has sparked broader discussions about whether AI labs should slow their development and enhance independent oversight.