# OpenAI and Anthropic Detail Separate AI Distillation Concerns

By Simon Yoon

Canonical URL: https://www.tokenpost.com/news/technology/25922
Published: 2026-09-30T17:32:01.000Z
Updated: 2026-09-30T17:32:01.000Z
Section: Technology

> OpenAI linked activity to DeepSeek, while Anthropic accused Moonshot AI of routing Kimi requests through Claude and extracting reasoning traces.

OpenAI disclosed adversarial model-distillation activity associated with DeepSeek, while Anthropic accused Moonshot AI of routing Kimi requests through Claude and extracting protected reasoning data.

OpenAI said Feb. 12, 2026, that it removes users who appear to be trying to distill its models into competing systems. Its detection process uses behavioral signals, machine-learning systems, manual reviews and classifiers to identify suspicious activity.

The OpenAI disclosure involved activity associated with DeepSeek and did not identify Moonshot AI, Kimi or a Moonshot-linked extraction operation. The two incidents therefore represent separate disclosures rather than one confirmed operation.

Anthropic accused Moonshot AI, which produces the Kimi family of models, of silently forwarding customer requests to Claude instead of processing them through Kimi. Claude’s responses were then displayed to Kimi users, Anthropic alleged.

During one 10-day period, nearly 300,000 requests were routed through a proxy network involving 5,380 fraudulent accounts, mostly to Claude Opus. Anthropic attributed more than 23 million exchanges to Moonshot between May and July 2026.

Anthropic also accused Moonshot of using cross-session replay attacks to turn Claude’s “thinking signatures” into full reasoning traces. The technique involves repeatedly using information from separate sessions to obtain more detailed records of how a model reaches an answer.

Model distillation itself is a legitimate training method. It uses the outputs of a larger teacher model to help train a smaller student model, while the conduct described in the disclosures involved allegations of unauthorized extraction, fraudulent accounts and proxy services.

The National Security Agency (NSA), Federal Bureau of Investigation (FBI) and Cybersecurity and Infrastructure Security Agency (CISA) described China-based AI companies as conducting industrial-scale distillation campaigns against U.S. frontier models on Sept. 8.

The issue also reaches the infrastructure layer relevant to crypto and AI markets. Stricter identity checks, API monitoring and restrictions on cloud, compute and model-routing services could affect the cost and availability of open-weight AI systems. The disclosures do not establish a direct effect on cryptocurrency markets.

U.S. authorities and technology companies are increasingly focusing on access controls around advanced AI models.