AI Labs Face New Third-Party Audits to Validate Safety Pledges

A new proposal for third-party auditing of leading AI labs, backed by Geoffrey Hinton, aims to institutionalize risk management by embedding evaluators directly within firms like OpenAI and Anthropic. This move formalizes the voluntary safety commitments made at the UK’s AI Safety Summit, shifting the industry from self-regulation toward a co-regulatory model. It directly addresses growing skepticism about the sincerity of corporate safety pledges, creating a framework to translate abstract principles into verifiable engineering practices and providing a crucial mechanism for accountability as model capabilities escalate toward AGI, a concern recently amplified by the disbanding of OpenAI