All news

OpenAI: reasoning extraction from GPT-6.2 Astra succeeds in 0.2% of attacks, versus 54.0% for GPT-6

Prepared by Sintelligent with the help of SI. Published in SI: . Updated: .

Rewriting an earlier message in the API discards every reasoning step generated after it, so a chain can no longer be assembled piecemeal from the conversation history.

On November 20 OpenAI disclosed a multi-layered defense against adversarial distillation — the coordinated siphoning of a model's outputs and reasoning chains in order to train a student model. According to OpenAI, a July campaign involved about 16,000 reasoning-extraction requests from more than 4,000 users. The main group, the company says, was in contact with people linked to Moonshot AI. In GPT-6.2 Astra and GPT-6.2 Sol, protected reasoning is tied to the conversation history. When a message is rewritten, the API discards all subsequent reasoning, and models from older families cannot read such chains. In an internal red team test, the success rate of extracting reasoning from a previous request fell from 54.0% for GPT-6 Astra to 0.2% for GPT-6.2 Astra. The protections include input and output classifiers, validation of tool-call arguments and account-level campaign analysis.

History

Dispute over distilling other labs' models

10 events
    1. OpenAI: reasoning extraction from GPT-6.2 Astra succeeds in 0.2% of attacks, versus 54.0% for GPT-6Open now
    1. Reuters: harvesting US model outputs in China has become industrial practice
    1. The Information: DeepSeek and Moonshot AI used GPT-6 Astra to train their latest models
    1. Anthropic says new measures cut data harvesting from Claude by competitors by 90-99%
    1. OpenAI says it stopped a campaign to "siphon off" its models' reasoning: the core is linked to people from Moonshot AI
    1. Jensen Huang: model distillation is "competition," not theft
    1. China opens a review of DeepSeek and Moonshot: according to Anthropic, they ran tens of millions of queries through Claude, including police data
    1. Y Combinator chief Garry Tan: American open labs should also be allowed to "distill" other models
    1. Anthropic: Alibaba, Moonshot and DeepSeek siphoned off Claude's knowledge through almost 200 million queries
    1. US intelligence agencies accuse DeepSeek, Moonshot, Alibaba and three other Chinese companies of mass "extraction" of American SI models