OpenAI accuses Moonshot AI of coordinated distillation campaign targeting GPT models

Moonshot AI-linked users fired 16,000 requests at OpenAI's models as part of a coordinated campaign to extract reasoning data, according to OpenAI¹⁴. The figure sharpens OpenAI's earlier accusation that individuals associated with the Chinese AI company orchestrated an industrial-scale model-distillation operation targeting its GPT systems⁴.

OpenAI said it disrupted the campaign, which involved mass data extraction from its AI models^1,2. The operation targeted hidden model reasoning, the internal chain-of-thought outputs that underpin OpenAI's reasoning-capable systems⁹. OpenAI has linked specific Moonshot AI individuals to the effort^3,10.

Model distillation is a technique in which a smaller or less capable model is trained to replicate the outputs of a larger one, effectively transferring knowledge without access to the original model's weights or architecture. At 16,000 requests, the volume OpenAI cites points to an industrial-scale operation rather than isolated probing.

Moonshot AI operates Kimi, a consumer-facing AI assistant^5,13. OpenAI's accusation amounts to a claim that Kimi's operator used OpenAI's production systems to harvest training signal for a competing product¹¹.

OpenAI said it blocked the large-scale extraction and disrupted the campaign. The company published its account under the title "Disrupting a coordinated model-distillation campaign".

ANALYSIS The 16,000-request figure, combined with OpenAI's decision to name Moonshot AI publicly, marks one of the most specific allegations of systematic model-output harvesting attributed to a rival lab.

A distillation campaign of this scale targeting production APIs raises questions about the adequacy of existing rate-limiting and output-monitoring defenses, particularly as OpenAI broadens access through agent-based products.

The accusation lands while OpenAI pursues a pre-IPO funding round of at least $30 billion at a roughly $1.4 trillion valuation^[3]. The reporting originates from Bloomberg and OpenAI's own published account. Moonshot AI's response has not been captured in the available evidence packet.