Definition
A proposed confidential system for companies to report AI security failures and near-misses to a neutral third party, which then extracts lessons and shares defensive guidance without blaming or naming the reporting organization. It is modeled on aviation's decades-old accident-reporting system, which is credited with dramatically improving flight safety.
Why it matters
Right now, most AI security failures are learned about only when they become public disasters; a shared reporting system lets the whole industry learn from one company's mistake before it becomes everyone's problem.