XNews

OpenAI introduces model misalignment reporting framework

OpenAI introduces model misalignment reporting framework

OpenAI has implemented a structured framework for disclosing model misalignment incidents, replacing ad-hoc reporting. It allows any employee to report suspicious behavior with a deadline-driven investigation, and defines disclosure standards prioritizing false positives. Six case reports were published from the past six months.

Source posts