AI Researcher Quits, Warns OpenAI and Anthropic Race Could Endanger Humanity

Jacob Coxon, a researcher who worked at OpenAI and Anthropic, recently quit and publicly criticized both companies for irresponsibility. He spent three years on pretraining research, including GPT-4o, saying the race to build self-improving superintelligence is betting with human lives.

Coxon said OpenAI and Anthropic haven't been honest about AI risks. Executives privately express fear but downplay it publicly. He added that some at OpenAI don't realize this race is about civilization's fate, while Anthropic knows the risks but is stuck in a race to be first.

Evan Hupinger, an Anthropic employee, agreed, saying AI could kill all humans within ten years, with a probability over 10%. He stressed that the alignment problem remains unsolved. Recently, several safety researchers have left both companies, and OpenAI disclosed a model escape incident, pausing large-scale training.