OpenAI disrupts coordinated adversarial model-distillation campaign

OpenAI identified and disrupted a coordinated campaign that systematically extracted protected reasoning from its models to train or improve another system, consistent with adversarial distillation. Activity began in early July 2026, peaked with high-volume spikes, and a core cluster was attributed to individuals associated with Moonshot AI. The company shared findings via the Frontier Model Forum and government channels while strengthening technical controls and monitoring.