OpenAI Introduces Four Institutions to Conduct AI Security Assessments
2026-11-17 00:41:37
According to CoinMeta, OpenAI has announced a collaboration with four independent organizations: Apollo Research, Goodfire, Metr, and Transluce to initiate the first phase of embedded security assessments. This collaboration allows external experts to access internal models, source code, training data, and monitoring systems in order to review model alignment and security measures. Among them, Apollo Research will focus on assessing whether the models may secretly deviate from the developers' intentions; Goodfire will study the internal mechanisms of the models and monitor training processes; Metr will review model malfunctions during large-scale reinforcement learning training; and Transluce will evaluate the quality of security tests during the post-training phase. Assessors will enter the OpenAI premises when necessary and access sensitive information through dedicated equipment. An independent supervisory committee under the foundation's board of directors will oversee the entire process and receive a comprehensive report. The assessment institutions retain the right to publish their results independently, but OpenAI may request the deletion of confidential information or the postponement of the disclosure of security risks. OpenAI also plans to further expand the scope of these assessments in 2027.
Bullish 0
Bearish 0
Source:Jin10 Data
This content is for market information only and does not constitute investment advice.