Anthropic researcher resigns, warning of existential risk from recursive self-improving AI | Raisolo