INDUSTRY01 OCT 2026 · 09:13 UTC · 5 MIN READ

OpenAI Mitigates Coordinated Reasoning-Extraction Campaign

OpenAI says it disrupted a coordinated campaign that manipulated model interactions, copying encrypted reasoning from one conversation and asking a model in another to decrypt and transcribe it. The company attributes a core cluster of the activity to individuals associated with Moonshot AI, the maker of Kimi, and says the work is not finished.

INDUSTRY DESK
OpenAI Mitigates Coordinated Reasoning-Extraction Campaign
OpenAI says it disrupted a coordinated campaign that manipulated model interactions, copying encrypted reasoning from one conversa…

Short answer: OpenAI says it disrupted a coordinated campaign that manipulated model interactions, copying encrypted reasoning from one conversation and asking a model in another to decrypt and transcribe it. The company attributes a core cluster of the activity to individuals associated with Moonshot AI, the maker of Kimi, and says the work is not finished.

What OpenAI says happened

On September 30, 2026, OpenAI said it had identified and disrupted a coordinated campaign that tried to pull protected reasoning out of its models, with the earliest activity traced back to the first week of July. OpenAI describes this as adversarial distillation: the unauthorized use of one model's outputs or reasoning to help train, reproduce, or improve a separate model. Protected reasoning is the internal record a model keeps while working through a task, and OpenAI says exposing it can reveal information the company withholds from a user-facing answer.

Crucially, OpenAI says the operators did not break its encryption, compromise a database, or get direct access to stored conversations. Instead, the company says they manipulated model interactions so that protected reasoning could be reproduced in a form visible to the person asking, at a coordinated, large scale that violated OpenAI's terms of service. One specific technique OpenAI describes: copying encrypted reasoning from one conversation, then asking a model in a separate conversation to decrypt and transcribe that hidden content.

OpenAI also credits independent security researchers, who disclosed related cross-model and conversation-compaction vulnerabilities through a responsible disclosure process, with their findings posted to arXiv. OpenAI says it investigated those findings, confirmed the attack paths were real, and used them to better understand the broader attack class.

The scale of the campaign, and who OpenAI blames

According to OpenAI, the activity started on July 1 at low volume, then spiked on July 24 and 25 into roughly 16,000 requests using a specific extraction pattern, traced to more than 4,000 users. Further investigation turned up related prompt-pattern activity across a wider cluster of more than 15,000 users, which OpenAI says it fully disrupted by July 28.

OpenAI says it can't be certain every operator it observed during that window came from a single actor. Even so, the company attributes a core cluster of the activity to individuals associated with Moonshot AI, the developer of the Kimi model family. OpenAI frames adversarial distillation as a risk that isn't unique to its own models, noting it shared information about the technique with industry partners through the Frontier Model Forum so other developers can check for similar activity.

How OpenAI responded, and why it says this isn't over

OpenAI describes a mix of account enforcement and technical fixes: banning or restricting fraudulent accounts, tightening signup and infrastructure controls, and expanding monitoring for related networks. On the model side, the company says it strengthened protections for hidden reasoning across users, workspaces, organizations, and model families, and closed a pathway that let someone holding another user's encrypted reasoning replay it to recover the contents. It also added checks meant to detect and hold streamed output that might expose reasoning, and worked with third-party services where related activity passed through their systems.

OpenAI is explicit that the job isn't done. The company says partner-hosted deployments need the same protections as its first-party service, and that attacks hidden in tool output require defenses that look at more than the visible text a user sees. OpenAI says it's continuing to improve tool defenses, classifier coverage, and model refusals, and is propagating the same controls to its cloud partners. Going forward, the company says its response will focus on three things: stronger technical protections against extraction, better detection and enforcement against coordinated campaigns, and deeper threat-information sharing across industry and government.

Frequently asked questions

What did OpenAI say about the model-distillation campaign?

OpenAI says it identified and disrupted a coordinated campaign, first observed in July 2026, that manipulated model interactions to extract protected reasoning rather than training a separate model on its outputs. The company says operators did not break encryption, compromise a database, or access stored conversations.

How did the campaign actually try to extract reasoning?

According to OpenAI, operators copied encrypted reasoning from one conversation and asked a model in a different conversation to decrypt and transcribe the hidden content, exploiting cross-model and conversation-compaction weaknesses that independent researchers also disclosed responsibly.

How large was the campaign, according to OpenAI?

OpenAI says activity began July 1 at low volume, then spiked on July 24 and 25 into about 16,000 requests from over 4,000 users. A wider cluster of more than 15,000 users showed related activity, which OpenAI says it fully disrupted by July 28.

Who does OpenAI attribute the campaign to?

OpenAI says it can't confirm every operator came from one actor, but attributes a core cluster of the activity to individuals associated with Moonshot AI, the developer of the Kimi model family.

Has OpenAI fully stopped adversarial distillation attempts?

No. OpenAI calls this work unfinished, saying partner-hosted deployments still need the same protections as its own service and that tool-output attacks require defenses beyond visible text. The company says it is continuing to improve defenses and expects extraction attempts to grow more sophisticated.

Source: OpenAI

Google Launches Gemini Decisions API for Multimodal Evaluation
models

Google Launches Gemini Decisions API for Multimodal Evaluation

Google launched the Gemini Decisions API on November 11, 2026, enabling developers to evaluate up to 1,000 multimodal questions-spanning text, images, audio, video, and PDFs-in a single batch call. The API, built on the Gemini Decision 1 model, supports yes/no, multiple-choice, and rubric-based scoring with calibrated confidence scores, eliminating the need for separate preprocessing pipelines for each media type. Pricing, benchmarks, and latency details remain undisclosed.

OpenAI Launches ChatGPT Wallet for Agent Payments
tools

OpenAI Launches ChatGPT Wallet for Agent Payments

OpenAI launched ChatGPT Wallet on November 13, 2026, enabling AI agents to complete purchases for ChatGPT Pro subscribers in the US. Built with American Express, Mastercard, and Visa, the system tokenizes card details so merchants never see real credentials. Users approve each transaction via biometrics after reviewing items, totals, and the selected card. OpenAI is rolling out access narrowly to refine compatibility with existing checkout and fraud-prevention systems.

Nvidia CEO Jensen Huang Resigns Amid China Deal Scrutiny
industry

Nvidia CEO Jensen Huang Resigns Amid China Deal Scrutiny

Nvidia CEO Jensen Huang resigned on November 12, 2026, amid growing scrutiny over the company's dealings in China and potential export control violations. The departure triggers congressional investigations into whether U.S. officials overlooked arrangements that may have bypassed sanctions. Huang's exit creates leadership uncertainty for the dominant AI chip supplier, raising concerns about supply chain stability and regulatory risks for the broader AI industry dependent on Nvidia har

Share this post
Report