Anthropic safety team member says he left over safety concerns, plans to join METR

@JoeJBenton said on September 11 that he had left Anthropic’s safety team two weeks earlier and would join METR to independently evaluate AI risks. He warned that AI development could pose extinction-level risks and argued that the public needs greater visibility into companies’ work. His proposed guardrails include disclosing progress toward recursive self-improvement—AI repeatedly improving itself—reporting safety incidents and near-misses, setting minimum safety standards and obtaining independent guarantees that companies meet them.