AI needs independent accident investigators

Following recent unexpected AI behaviors at OpenAI and Google, experts are calling for independent accident investigators for AI systems, moving beyond self-reporting by the companies that build them. This push follows an open letter signed by over 100 AI experts. Currently, the companies that create AI systems also control the disclosure and framing of incidents, creating a conflict of interest. Proponents suggest a model similar to aviation, where independent bodies investigate after a major failure, requiring full transparency and access to training records and incident logs. To address this, Rumman Chowdhury launched the Independent AI Evaluation Foundation (IAEF) at the U.N. General Assembly with $10 million in funding. The foundation aims to professionalize independent AI evaluation, create an open-access testing environment, and establish a post-incident investigation loop that currently does not exist.