OpenAI has disrupted a campaign linked to Moonshot AI aimed at mimicking its AI's internal reasoning, reporting thousands of unauthorized extraction attempts.
OpenAI Discovers Extraction Campaign
OpenAI has announced the disruption of a coordinated effort aimed at mimicking the internal reasoning of its AI models, which is suspected to involve individuals associated with Moonshot AI, the developer behind the Kimi chatbot.
This campaign began on July 1 and resulted in 16,000 extraction attempts from over 4,000 users on July 24 and 25 alone, part of a larger group exceeding 15,000 users. OpenAI claims to have successfully halted these activities by July 28.
Objective of the Campaign
The main goal of the initiative was to access the internal reasoning processes rather than simply retrieving answers from the AI. OpenAI emphasizes that its models engage in complex reasoning before producing outputs, and unauthorized access could reveal sensitive information not included in final responses.
“The operators did not breach our encryption, compromise any databases, or gain direct access to stored user conversations,” OpenAI stated.
Instead, the attackers manipulated interactions with the models to expose protected reasoning, which constitutes a violation of OpenAI's terms of service. Techniques included extracting encrypted reasoning from one interaction and then requesting the model to decode it in a different context.
Concerns Over Distillation
OpenAI highlighted that the unauthorized extraction of internal reasoning is part of a larger issue known as adversarial distillation, which involves using one model's outputs to enhance another without permission. This practice raises significant concerns, especially as distillation can lead to the development of smaller models that perform comparably to larger ones without extensive training.
The company has taken steps to close vulnerabilities that allowed such manipulations and continues to monitor for similar activities.
Broader Context of AI Controversies
This incident is not isolated; OpenAI has faced similar allegations previously. In early 2025, the company investigated potential distillation attempts by DeepSeek amid national security concerns. Other AI firms, including Anthropic, have reported similar issues, with accusations against Chinese entities employing fake accounts to interact with their models.
As of now, Moonshot AI has not publicly responded to these allegations and is currently pursuing a $3 billion IPO in Hong Kong, aiming for a valuation of $50 billion.
