OpenAI says Moonshot AI ran 16,000-request campaign to extract model reasoning

September 30, 2026 at 11:27 PM2 min

This digest was compiled by AI from multiple sources — links to the originals are below.

OpenAI disrupted a coordinated campaign that made 16,000 requests on July 24-25 to extract protected reasoning from its models. The company attributed a core cluster of the activity to individuals associated with Moonshot AI, developer of the Kimi model. The campaign began July 1 and was fully disrupted by July 28.

Key Facts

OpenAI observed the first activity in the first week of July, with high-volume spikes of 16,000 requests on July 24 and 25 from over 4,000 users.
The prompt-pattern activity was spread across a cluster of more than 15,000 users.
OpenAI fully disrupted the campaign by July 28.
OpenAI attributed a core cluster of the activity to individuals associated with Moonshot AI, the developer of Kimi.
Anthropic has made similar accusations against Moonshot AI.

Campaign Mechanics

The campaign copied encrypted reasoning from one conversation and asked a model in another conversation to decrypt and transcribe the hidden reasoning content. OpenAI said the activity began on July 1, initially at a low volume, before high-volume spikes on July 24 and 25 consisting of 16,000 requests using a relevant extraction pattern from over 4,000 users. The prompt-pattern activity was spread across a cluster of more than 15,000 users. OpenAI was able to fully disrupt the campaign by July 28.

Attribution and Response

OpenAI attributed a core cluster of the activity to individuals associated with Moonshot AI, the developer of Kimi. The company has not linked the activity to a single actor. Anthropic has made similar accusations against Moonshot AI. In late July, Michael Kratsios, Director of the White House Office of Science and Technology Policy, declared that the US government had information that Moonshot distilled its Kimi K3 model from Anthropic's models.

Research Findings

A study found that feeding a 1 percent fragment of Claude Opus 4.8's decoded reasoning into Kimi K3 shifted its subsequent thinking and final answer to match Claude's style and wording. The study authors reported that specific Claude and GPT reasoning spans are up to ~6 orders of magnitude easier to extract from Kimi-K3 than from the next-closest model. Kimi K3 exhibited an unusually high statistical probability of predicting and continuing Claude's text compared to other open models like DeepSeek-V4-Flash.