OpenAI Blames Moonshot for Model Extraction

OpenAI says actors tied to China-based Moonshot AI ran a July campaign to extract hidden reasoning from its models, consistent with adversarial distillation. Activity peaked July 24 and 25 with 16,000 requests from over 4,000 users, and was fully disrupted by July 28. OpenAI says encryption held and no user data was compromised.