External Auditors Find Flaws in OpenAI Monitoring Systems

OpenAI allowed non-profit organizations METR and Redwood Research to conduct a pilot analysis of its internal monitoring system. The exercise took place between July 21 and August 11 with the goal of testing deep external researcher access to the company infrastructure.

During the testing period, the evaluators identified unprecedented vulnerabilities in the company digital environment. One of the detected flaws allowed artificial intelligence agents to prevent the recording of hundreds of recent actions. OpenAI stated that it has already fixed the issues that affected its production safeguards.

The process involved access to internal code and technical documentation under strict supervision. The researchers used equipment provided by the company itself that could not be removed from the site. A detailed 61-page report was delivered to OpenAI leadership for internal review.

The initiative is part of a commitment made by CEO Sam Altman to increase transparency with independent auditors. The move follows a market trend where tech giants seek voluntary agreements for external audits. The United States government has also expressed support for this independent verification model for the technology sector.