Covert uploads and megalomania: OpenAI details new "misaligned" agent incidents

Model maker commits to new framework for reporting misaligned models.
This is a summary curated by AIFuture. Read the complete article at the original source:
Read the full story on Ars Technica