Jacob Coxon, a researcher at Anthropic, announced his resignation and accuses Anthropic and OpenAI of racing toward self-improving superintelligence at the expense of human safety. He claims to have spent three years working on pre-training at both companies and argues that upcoming AI systems will be capable of hacking any system, transforming any field overnight, and acquiring real power. According to him, private conversations within the industry sharply contrast with public messaging, with senior officials softening their language for the press while expressing fears privately. Over 1,100 frontier lab staff signed the « Pacing the Frontier » letter in July, requesting that the U.S. government support an international effort to deliberately slow automated AI development.
Source: Read the original article
