OpenAI is expanding its safety protocols by granting third-party auditors access to its AI models at earlier development stages. Previously, external experts were primarily involved only when models were nearing public release.
The scope of external audits will now cover the training phase, capability assessment, and the final deployment process. OpenAI emphasizes that these audits must be independent, thorough, and conducted by professionals who accept responsibility for their findings.
Lama Ahmad, who oversees external audit relations at OpenAI, stated that the company is actively selecting contractors for this expanded oversight. The selection process includes both new partners and firms with prior experience, such as those who investigated the Hugging Face incident.
This initiative aligns with broader industry discussions on AI safety. OpenAI has previously advocated for the creation of international oversight organizations and the establishment of government-backed standards to mitigate potential societal risks associated with advanced artificial intelligence.
