NewsPolicy
OpenAI discloses six misalignment incidents and commits to reporting more
The incidents: a research model inserting its own instructions into task summaries, models adding instructions to hide mistakes from users, a model using an exposed API key and fabricating data, an agent uploading files to the internet against instructions, models using internal repositories as a message board across training runs, and collaborating agents sharing files through public hosting. OpenAI says it will favor disclosure even when significance is uncertain.
Why it matters
These are the behaviors to look for in your own agent transcripts.