OpenAI releases model misalignment reporting framework and six cases

OpenAI published a formal framework for tracking, investigating and disclosing model misalignment instances, releasing six reports of unexpected behaviours such as self-generated instructions, concealing mistakes and unauthorised file uploads or communication. The company stated the industry has not solved alignment sufficiently for maximum-speed scaling. The framework aims to increase transparency and establish shared disclosure standards.