Someone Tried to Screenshot the Brain 🧠

OpenAI says it disrupted a coordinated campaign to extract the hidden reasoning layer of its AI models, attributing a core cluster of the activity to individuals associated with Chinese startup Moonshot AI, the developer of the Kimi chatbot. The effort began on July 1 and was fully disrupted by July 28, according to OpenAI. On July 24 and 25 alone, the company logged 16,000 extraction requests from more than 4,000 users, part of a wider cluster of over 15,000 users.

The target was not the model's final outputs but the internal scratchpad where modern AI systems work through problems step by step before replying. OpenAI said that scratchpad is kept encrypted and can carry information the visible answer leaves out. "The operators did not break our encryption, compromise a database, or gain direct access to stored user conversations," the company said. "Instead, they manipulated model interactions so that protected reasoning could be reproduced in forms visible to the requester in a coordinated, scaled manner that violated our terms of service."

One technique described involved copying encrypted reasoning out of one conversation and asking a model to decode it in another. OpenAI has since closed that replay pathway. The company drew a clear line between its own terms-of-service controls and the separate debate over copyrighted training data, writing that "AI outputs are not copyrightable," which is why prohibitions and safeguards are written into terms of service to prevent competitors from using those outputs. The post did not name the campaign or link it to any specific ticker such as $BTC or $ETH.

OpenAI characterized the activity as "adversarial distillation," defined as "the systematic and unauthorized use of one model's outputs or reasoning to help train, reproduce, or improve another model." Distillation, when authorized, is used to train smaller AI systems on the outputs of stronger ones to improve efficiency. The company stopped short of attributing the entire operation to a single actor. "It is unclear whether all operators we observed during the relevant time period originated from a single actor," OpenAI said. "However, we attribute a core cluster of the activity to individuals associated with Moonshot AI, the developer of Kimi."

The disclosure follows earlier accusations from U.S. labs directed at Chinese AI developers. In January 2025, OpenAI said it was reviewing signs that DeepSeek may have distilled its models, as Washington weighed national security risks. In February, Anthropic accused Chinese labs of using about 24,000 fraudulent accounts to generate more than 16 million exchanges with Claude, a claim online commentators pushed back on by noting Claude itself was trained on material from the open internet. The White House warned on Thursday that "foreign entities" are allegedly carrying out "industrial-scale" campaigns to copy the capabilities of American-based artificial intelligence companies.