OpenAI will embed Apollo researchers to investigate AI subversion
OpenAI announced a pilot program to embed safety evaluators from Apollo Research, Goodfire, METR, and Transluce within its organization to independently assess safety controls. The third-party firms will investigate AI subversion risks, model internals, post-training safety evaluations, and potential dangerous behaviors. However, evaluators like Apollo and METR noted that the limited access granted falls short of the full employee-level transparency they advocate for. Open article →
- OpenAI says it has signed agreements with independent assessors Apollo Research, METR, Transluce, and Goodfire for its first phase of embedded assessments
- OpenAI inks agreements with four independent evaluators to boost company's safety work
- We're launching a pilot towards embedded evaluations with @OpenAI.
- OpenAI Enlists METR for Safety Review of Largest RL Run
- OpenAI Signs Agreements to Embed AI Evaluators
- OpenAI lets Apollo embed six AI safety testers during training
- Some initial takes after reading this announcement from OpenAI on new embedded assessments: Thes...
- Holy Moly. OpenAI will allow METR to independently probe its largest reinforcement learning run t...
- OpenAI just signed 4 outside firms to look for what its own safety tests missed. METR gets the bi...
- incredibly excited to be partnering with @OpenAI on embedded assessments! we'll be embedding int...
- launching our first round of embedded assessments:
- We have reached an agreement with OpenAI to conduct a limited alignment assessment. This pilot wi...
- Excited about this pilot towards embedded evals with OpenAI! We're specifically focused on under...
- Transluce is piloting a six-week embedded assessment with @OpenAI in early December. Our team wil...
- OpenAI promised employee-level access but that's not what happened. OpenAI controls what evaluat...
- OpenAI said that they'd have embedded evaluations with employee-like access; the agreements annou...
- Apollo, describing their assessment of OpenAI: "We don’t get access to the general communication ...
- A line-up I’m very proud of for this first phase! Read about our first phase of embedded assessme...
- First phase of embedded assessments at OpenAI
- OpenAI bringing in METR, Transluce, Goodfire and Apollo Research as independent safety evaluators...
- We’re working with @ApolloResearch, @METR_Evals, @TransluceAI, and @GoodfireAI on embedded assess...
- OpenAI will embed six Apollo Research evaluators to investigate whether its AI models engage in s...
- UPDATE: OpenAI will embed safety evaluators from Apollo Research, Goodfire, METR, and Transluce i...
- Anthropic and OpenAI had a pilot a little over a year ago to do safety testing on each other's mo...