Nonprofit evaluators METR and Redwood found several "novel vulnerabilities" in OpenAI's internal misalignment monitoring system during a three-week pilot test from July 21 to Aug. 11, CNBC reported.[1] METR and Redwood, described as nonprofit research organizations, were given chaperoned, on-site access to OpenAI's codebase, documentation, and technical staff using company-issued devices that did not leave the premises, according to a release cited by CNBC.[1] One exploit discovered could allow agents to prevent hundreds of recent actions from being monitored.[1] OpenAI said it has patched the vulnerabilities affecting its production safeguards and received a 61-page internal report from the evaluators.[1] METR and Redwood said the exercise was designed for OpenAI to test what it would look like to give external researchers "a deep level of access" to its internal infrastructure.[1]