OpenAI reasoning-extraction campaign

What OpenAI disclosed

OpenAI said on September 30 that it had identified and disrupted a coordinated campaign to extract protected reasoning from its models. The company calls the activity adversarial distillation: attempts to use another model’s hidden reasoning or outputs to reproduce capabilities. OpenAI says the earliest observed activity began in July and that the related account cluster was disrupted by July 28. These are findings from the company’s own investigation, not an independently verified account of every operator’s activity.

Distinct from a database breach

OpenAI says the operators did not break encryption, compromise a database or directly access stored user conversations. It describes attempts to make protected reasoning visible through model interactions, and says it responded with account enforcement, technical controls and information sharing with industry partners. OpenAI reports large numbers of attempted requests but notes that attempts should not be confused with successful extractions. The company says mitigation work continues, including protection across partner-hosted deployments.