OpenAI Grants Apollo, METR, Transluce and Goodfire On-Site Access to Model Code
Summary
OpenAI signs agreements with Apollo Research, METR, Transluce and Goodfire for a first phase of embedded assessments. The assessors receive model checkpoints, source code and training data, and work on site at OpenAI offices on company devices. METR will examine OpenAI's largest reinforcement learning run to date.
Key Points
- Apollo Research tests four claims across two stages, including whether training incentivized subversion, OpenAI says.
- OpenAI says Apollo Research and Goodfire accept financial compensation that is not contingent on assessment results.
- The Safety and Security Committee, an independent oversight committee of the OpenAI Foundation's Board, receives unredacted reports.