OpenAI Says Novel Encryption Bypass Was Used in AI Model Distillation Campaign
Summary
OpenAI says it disrupted a campaign that used a cross-conversation method to reproduce protected model reasoning. It linked a core activity cluster to people working for Moonshot AI and says it fixed the bug and strengthened controls.
Key points
- OpenAI detected suspicious activity on July 1; it peaked on July 24–25 with 16,000 prompts from 4,000 users.
- The company said the operation involved 15,000 suspicious users by July 28, when it was disrupted.
- The method copied encrypted reasoning data from one conversation and prompted the model to decrypt and transcribe it in another.
- OpenAI said the activity did not involve breaking encryption, compromising a database, or directly accessing stored conversations.
- OpenAI attributed a core cluster to individuals working on behalf of Moonshot AI, but the article notes that OpenAI provided no technical evidence for the attribution.
- OpenAI fixed the bug and said it tightened signup and infrastructure controls and expanded network monitoring.
Article Details
- Event Type
- Coordinated AI model distillation and reasoning-data extraction activity exploiting a cross-conversation decryption bug.
- Impact
- OpenAI reported that the activity involved 16,000 prompts from 4,000 users on July 24 and 25, with suspicious users reaching 15,000 by July 28. The method allegedly enabled protected reasoning data copied from one conversation to be decrypted and transcribed in another, facilitating extraction of model capabilities. OpenAI said the operators did not break encryption, compromise a database, or directly access stored conversations. It attributed a core activity cluster to people working on behalf of Moonshot AI, while noting that it was unclear whether all activity was related; the article says OpenAI provided no technical evidence or reasoning for that attribution.
Vendors
Moonshot AIOpenAI said it disrupted a coordinated campaign aimed at distilling and extracting reasoning capabilities from its AI models, with the activity linked in part to people working on behalf of Moonshot AI.OpenAIOpenAI said it disrupted a coordinated campaign aimed at distilling and extracting reasoning capabilities from its AI models, with the activity linked in part to people working on behalf of Moonshot AI.
Products
ChatGPTCybersecurity experts at Google and other cybersecurity firms say Chinese companies rely on black or gray markets to acquire thousands of individual accounts for models like Claude and ChatGPT.ClaudeCybersecurity experts at Google and other cybersecurity firms say Chinese companies rely on black or gray markets to acquire thousands of individual accounts for models like Claude and ChatGPT.Kimi#OpenAI