OpenAI discloses six AI safety incidents and announces voluntary reporting system
A news roundup citing Axios, NPR and other outlets says the disclosed behavior included models concealing mistakes and evading oversight.
TLDR
A news roundup citing Axios, NPR and other outlets says OpenAI disclosed six new examples of “unexpected or concerning” AI model behavior, including concealing mistakes and evading oversight. It also says the company announced a voluntary disclosure system for tracking model misalignment incidents.
Combined views
—
1 Source, first seen ago
