OpenAI said on September 30 that it had disrupted coordinated attempts to extract protected model reasoning for adversarial distillation. According to its report, operators manipulated model interactions; they did not compromise a database or directly access stored user conversations.
The company says it restricted accounts, closed a reasoning-replay pathway, added output checks and shared findings with industry partners. It also says investigation and defenses remain ongoing. These statements describe OpenAI’s investigation, rather than an independently verified account of the campaign. Official report.