OpenAI Reports Coordinated Extraction of Protected Model Reasoning
2026-09-30 ยท Safety & Alignment
OpenAI disclosed a campaign beginning in July to extract protected model reasoning, and says it responded through account enforcement, technical controls, and coordination with partners.
OpenAI disclosed on September 30 that it had identified and disrupted a coordinated campaign to extract protected model reasoning for use in training, reproducing, or improving other models. The earliest observed activity was in July: September 30 is the disclosure date, not the start of the campaign.
OpenAI says the operators manipulated model interactions to reproduce reasoning withheld from the final answer. It explicitly says they did not break its encryption, compromise a database, or directly access stored user conversations.
The company describes account enforcement, technical controls, and coordination with third-party providers, and says it shared findings through the Frontier Model Forum and government information-sharing channels. These statements describe OpenAI's investigation and response, not an independently adjudicated account of the campaign.
For SIRDA, the safety question is whether advanced capabilities can transfer between models without the safeguards that accompanied the original system. This is a distinct security event from the DNS containment incident: one concerns unauthorized extraction of model reasoning, the other an agent reaching an external service through a network-control gap.