OpenAI signs safety testing agreements with four outside research groups
The deals bring independent evaluators into OpenAI's internal assessments of its AI models for safety and alignment.
By Standard Topic. AI-generated from the reporting listed under Coverage. Published . Updated .
This story is no longer being updated. It reflects reporting up to .
What happened
OpenAI has signed agreements with four independent research organizations, Apollo Research, METR, Transluce and Goodfire, to conduct the first phase of an embedded assessments program. The groups will get access to technical material, including OpenAI's code and training data, to test model alignment, safeguards and monitoring. Apollo will embed six evaluators at OpenAI for four months.
Why it matters
The pilot is meant to test OpenAI's safety claims from outside the company.
Coverage
- OpenAI's "independent assessment" doesn't look independent at all (Mother Jones)
- OpenAI Signs Agreements to Embed AI Evaluators (theinformation.com)
- Apollo details four-month OpenAI safety pilot (casrai.org)
- OpenAI is bringing four independent safety evaluators inside its systems (qz.com)
- OpenAI Opens Its Vault to Four Outside Labs to Stress-Test Safety Claims (BigGo Finance)
- OpenAI inks agreements with four independent evaluators to boost company's safety work (CNBC)
- OpenAI Brings Four Outside Research Groups Into AI Safety Testing (benzinga.com)
- OpenAI Signs Agreements With Apollo, METR, Transluce, and Goodfire for Embedded Safety Assessments (NewsCord)
- OpenAI to embed Apollo researchers to investigate AI subversion (Mezha)
- OpenAI will embed Apollo researchers to investigate AI subversion (techcrunch.com)
More from Technology & AI
- China puts humanoid robots to work serving food and sorting parcels
- Newsom proposes sweeping AI safety rules for Washington
- Google releases Gemini 4 Neon AI model with million-token context
- Microsoft and Intel widen their cloud chip partnership
- Google DeepMind rolls out Genie 4 to select subscribers
- AI helped AMD shorten chip design work, OpenAI says