News · curated 30 Sep 2026
Disrupting a coordinated model-distillation campaign
First reported openai.com
Coverage timeline
Single-source incident — first reported, latest, and curated coincide.
Why it matters
Adversarial distillation lets rivals reproduce a frontier model's capabilities and reasoning while stripping its safety guardrails, and the reasoning-extraction techniques described (cross-model and conversation-compaction prompt patterns) represent a broader attack class affecting many LLM providers.
OpenAI reported disrupting a coordinated adversarial model-distillation campaign it attributes to individuals associated with China's Moonshot AI (developer of Kimi), which ran through July 2026. The operators manipulated model interactions to reproduce protected internal reasoning at scale — including copying encrypted reasoning from one conversation and asking a model in another to decrypt and transcribe it — with spikes of 16,000 requests across 4,000+ users on July 24-25 and related prompt-pattern activity across 15,000+ accounts. OpenAI says no encryption was broken or database compromised, and it enforced mitigations and account actions by July 28.