OpenAI reveals ‘novel’ encryption bypass used in distillation attack
OpenAI disrupted a Moonshot AI-linked distillation campaign that tricked its models into decrypting protected reasoning data using up to 15,000 accounts.
OpenAI said it disrupted a coordinated campaign that copied encrypted reasoning data from one conversation and asked the model in a separate conversation to decrypt and transcribe it as plaintext. Activity peaked at 16,000 prompts from 4,000 users on July 24-25 and reached 15,000 suspicious users before full disruption on July 28. OpenAI attributed a core cluster to individuals working on behalf of Chinese rival Moonshot AI but published no technical evidence. The company says the same vulnerability exists in other AI models, shared details with the Frontier Model Forum, and fixed the cross-conversation decryption bug.
- Attackers copied encrypted reasoning data and used separate conversations to decrypt it
- Attributed to individuals working for China's Moonshot AI without published technical evidence
- Scaled to 15,000 suspicious users and 16,000 prompts before July 28 disruption
- Same vulnerability reportedly exists in other AI models; details shared with Frontier Model Forum
- Organizations
- OpenAIMoonshot AIFrontier Model Forum
- AI models
- Kimi
From the original article
3 min read at cyberscoop.com
The company said individuals associated with Chinese company MoonshotAI were behind parts of the attack, but did not offer hard evidence for the claim. The post OpenAI reveals ‘novel’ encryption bypass used in distillation attack appeared first on CyberScoop.
By djohnson · cyberscoop.com
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.