An article summarized by Axios:
Three Anthropic researchers have publicly raised serious concerns about the potential risks of advanced AI, warning that AI could potentially cause catastrophic harm to humanity within the next decade. Jacob Coxon, who resigned from Anthropic, said AI developers privately believe the technology could “kill us all,” while alignment researcher Evan Hubinger agreed and said he personally puts the chance of human extinction from AI at more than 10% over the next decade. Another Anthropic researcher, Samuel Marks, said these concerns are increasingly common among senior AI employees.
The warnings highlight a growing dilemma inside the AI industry: slow development to reduce safety risks or continue racing ahead and risk losing control of increasingly powerful systems. The researchers argue that AI capabilities are improving faster than expected and could eventually reach a point where systems can significantly improve themselves, making their behavior increasingly difficult to predict or control.
The comments are part of a broader debate over AI safety, regulation and the pace of development. Critics argue that AI companies may have incentives to emphasize existential risks because stricter regulation could strengthen the position of established firms, while supporters say the concerns should be taken seriously because researchers have access to increasingly powerful systems that the public has not yet seen. Importantly, the researchers are not saying human extinction is imminent; they are warning that the risk could become much more serious as AI capabilities continue advancing.
