A post proposes a 'model card' equivalent for reporting AI misalignment incidents. Suggested fields include dates, frequency, whether behavior occurred during evaluation or reinforcement-learning training, whether monitoring caught it, task category, model family, novelty and external impact. The author argues that standardized reporting would improve transparency and understanding, but stresses that it must not slow disclosure. Quoting the proposal, a researcher says their team has already built such a system and shares links to a paper and demo. They are seeking an independent organization to maintain a registry and follow up with model providers, saying individual researchers lack the bandwidth.