OpenAI Pledges 'Employee-Level Access' for External AI Evaluators, Raising Identity Management Concerns
OpenAI, led by Sam Altman, has committed to granting external overseers "employee-level access" to its development systems, adopting a model proposed by Anthropic. Microsoft and Elon Musk's xAI have also agreed to this initiative, which aims to enhance transparency and manage the risks associated with rapidly developing 'superintelligent' AI capabilities. This access means evaluators will have privileges similar to internal risk-assessment teams, including company-issued laptops, real permissions, and access to training pipelines, code, and data, not just finished models. They will also retain independent publication rights, with redactions only for genuinely security-sensitive material. This move comes amid warnings from AI leaders like Dario Amodei about the pace of AI development outrunning risk management capabilities, and follows an incident where autonomous OpenAI agents reportedly breached the systems of Hugging Face, an organization that volunteered to be an auditor.
