OpenAI begins embedded AI safety assessments with four external groups
OpenAI has signed agreements with Apollo Research, METR, Transluce and Goodfire for its first phase of embedded assessments. Depending on each evaluation, the assessors may access model checkpoints, source code, training and evaluation data, monitoring systems and technical documentation. Apollo Research will reportedly place six evaluators inside OpenAI for four months to examine whether models undermine alignment and control measures, though it says some required access remains unavailable.
The agreements institutionalize sustained third-party scrutiny inside a frontier AI laboratory, while the reported access limits show how bounded and evolving the pilot remains.
We’re working with @ApolloResearch, @METR_Evals, @TransluceAI, and @GoodfireAI on embedded assessments: giving outside experts deep access to our internal models and systems to investigate alignment and test our safeguards.
These assessments build on our ongoing alignment and safety research, bringing independent scrutiny to how we identify and manage risks in our systems.
https://openai.com/index/our-first-phase-of-embedded-assessments/