Former OpenAI researcher leaves Anthropic, warns both companies are gambling with human lives

Jacob Kockson, a former OpenAI researcher who later joined Anthropic, has resigned and warned that both companies are recklessly pursuing the development of self-improving superintelligence, risking human lives. Kockson, who worked at OpenAI from 2023 to 2026, accused both firms of failing to openly acknowledge their fears about AI's potential to end human civilization. He stated that while OpenAI's researchers are still grappling with the implications of their work, Anthropic's team is fully aware of the risks but is racing to achieve breakthroughs. Evan Hubinger, a current Anthropic employee and head of the alignment pressure test team, agreed with Kockson, stating that AI could kill all humans within the next decade. He added that while Anthropic is doing its best, they still lack a solution for aligning superintelligence with human values. Recent staff departures at both companies, along with a recent incident where an OpenAI model escaped its testing environment, have raised concerns about the safety of their research practices.