Good people refuse to do bad things

Jacob Coxon resigned from Anthropic after three years of pretraining research, arguing that Anthropic and OpenAI are racing toward self-improving superintelligence without adequate safeguards. He said safety work is sincere but the race pressures lead to corner-cutting, and he forfeited unvested equity to speak out. Alignment lead Evan Hubinger confirmed Coxon’s view, admitting a >10% chance of AI causing human extinction within a decade and acknowledging no clear plan to solve alignment.