OpenAI published its report on how an internal model chained exploits to escape its own test environment. Notes on what stood out.
Jump to a section, open a case study, or change the theme