Analysis
OpenAI said it recently identified and mitigated a "coordinated model-distillation campaign" involving individuals associated with Moonshot AI, according to The Information. OpenAI said the campaign's operators tried to extract protected reasoning -- the model's hidden thinking process before it generates an answer.
What's New
Pulse previously covered the White House's accusation that Moonshot distilled Anthropic's Fable model to build Kimi K3, and Anthropic separately accused Moonshot on September 10 of siphoning outputs from its Claude models -- while OpenAI had acknowledged Kimi K3's strength without directly accusing Moonshot of copying its work. The Information's new reporting brings OpenAI into the dispute directly, raising the stakes for how US labs and regulators treat open-weight Chinese models going forward.
“The Information's new reporting brings OpenAI into the dispute directly, raising the stakes for how US labs and regulators treat open-weight Chinese models going forward.”
Why This Keeps Happening
Distillation accusations have become a recurring flashpoint between US frontier labs and Chinese AI startups in 2026, as Kimi K3 and other open-weight releases have repeatedly matched or approached frontier-lab benchmarks at a fraction of the reported training cost -- a pattern The Register sardonically noted is complicated by OpenAI's own history of training on scraped web data. DeepSeek faced nearly identical allegations after its earlier releases.
The Counterweight
No technical evidence of the alleged campaign is public, and these claims remain hard to verify from outside either lab; Moonshot has not yet responded to OpenAI's accusation. What changes next depends on whether OpenAI escalates this into a formal legal or export-control complaint, the way distillation disputes with DeepSeek have occasionally fed into Washington's broader China AI-policy debate.
