OpenAI reports 6 new instances of 'concerning model behavior' since March
2026-09-16 · Tech
OpenAI graciously admitted to six new cases of their models behaving badly since March, because apparently one public meltdown at a time wasn't enough. They even built a whole framework for disclosing future screwups — optimistic planning for a company that keeps finding new ways to ship concerning behavior. At least they're getting efficient at something: cataloging their own failures.