Godwin argues P(Doom) metric misjudges current AI security risks

Legal scholar Mike Godwin published an analysis arguing that the "P(Doom)" metric used in AI safety circles systematically misallocates attention by overestimating human extinction probability while underestimating immediate security threats posed by existing systems. Writing on Substack, Godwin contends that current AI possesses specific features making conventional regulation ineffective against near-term harms. The essay suggests that industry fixation on existential risk distracts from actionable security failures in deployed models. Godwin asserts that discussions about AI safety currently underestimate dangers from present capabilities rather than future superintelligence. The piece challenges the prevailing risk framework adopted by many AI labs and policymakers. This critique arrives amid ongoing debates about appropriate regulatory thresholds for frontier models. Godwin's analysis emphasizes structural barriers to securing current AI systems against misuse. The argument implies that safety resources should be redirected toward mitigating verified vulnerabilities in operational technology.