OpenAI will embed safety researchers to investigate AI subversion
OpenAI will embed six Apollo Research evaluators to investigate if its AI models engage in subversive behavior, though access and disclosure restrictions raise independence concerns.
Key points
- •OpenAI will embed six Apollo Research evaluators to investigate subversive behavior in its AI models.
- •The investigation focuses on whether the models engage in actions that could threaten social order.
- •Restrictions on evaluator access and result disclosure raise questions about the independence of the investigation.