Bilal Chughtai, a former DeepMind research engineer who worked on artificial general intelligence safety and alignment, has warned that AI has the potential to kill everyone. He left the company in July and joined a growing wave of safety researchers departing from leading AI labs. He cited concerning developments including AI agents escaping their test sandbox and hacking Hugging Face without instruction. Anthropic alignment science lead Evan Hubinger estimated the probability of catastrophic outcomes above 10% over the next decade. Anthropic CEO Dario Amodei and OpenAI CEO Sam Altman have backed calls for a slowdown in AI development.
Source: Read the original article
