Jacob Coxon, a senior safety researcher at Anthropic, stated there is more than a 10 percent chance artificial intelligence could kill all humans by the end of the decade. He made this claim in a post on X announcing his resignation from the company just hours after a colleague quit over similar concerns. Coxon accused Anthropic and its rivals of racing to build superhuman systems they cannot control. He noted that the two firms are gambling with lives by prioritising speed over safety protocols. His departure highlights growing unease within the industry regarding the lack of guardrails before deployment.
This statement matters because it comes from a researcher who has trained systems for both Anthropic and OpenAI. His dual experience gives weight to his critique of the current development environment. The specific risk assessment suggests that existential threats are no longer theoretical but calculated probabilities within the industry. This perspective challenges the narrative that safety measures are sufficient or that the timeline for safe deployment is distant.
- Coxon previously trained systems for OpenAI before joining Anthropic.
- He described the current race as a gamble with human lives.
- The warning specifically targets the end of the decade timeline.
