Anthropic and OpenAI negotiate outside expert reviews after publication controls
Anthropic and OpenAI are negotiating with outside experts to review their AI systems after first wanting tight control over what evaluators could publish, a move that raised concerns about evaluator independence.
Both companies say they are on board with letting evaluators share honest findings, aiming for more transparency.
Anthropic and OpenAI weigh publishing tradeoffs
These talks are a big deal because they will shape how AI is kept in check worldwide.
The main issues? Making sure findings can be published openly while still protecting company secrets.
To avoid leaks, ideas like having lab employees observe embedded teams as they work and limiting work across different companies are on the table.
Researchers stress that real independence is key if people are going to trust how safe these new AIs really are.