Disrupting a coordinated model-distillation campaign

OpenAI announced the disruption of a coordinated campaign designed to extract protected reasoning from its models through adversarial distillation, a process where one model's outputs or internal reasoning are used to train or improve another. The earliest observed activity occurred during the first week of July, followed by high-volume spikes on July 24 and 25 that involved thousands of requests and users. OpenAI fully disrupted the related prompt-pattern activity across a cluster of more than 15,000 users by July 28. While it remains unclear if all operators originated from a single actor, OpenAI attributed a core cluster of the activity to individuals associated with Moonshot AI, the developer of Kimi. According to OpenAI, the operators did not breach encryption or compromise databases, but instead manipulated model interactions to make protected reasoning visible to requesters. Independent security researchers also brought cross-model and conversation-compaction vulnerabilities to OpenAI's attention, accelerating mitigation efforts. To combat the campaign, OpenAI deployed account enforcement, banned fraudulent accounts, strengthened signup and infrastructure controls, closed pathways that allowed the replay of encrypted reasoning, and added checks for streamed outputs. OpenAI also shared its findings with industry partners through the Frontier Model Forum and government channels, noting that adversarial distillation presents shared safety and national security risks across the industry by potentially bypassing original model safeguards.