Analysis
OpenAI has told Moonshot AI it believes the Chinese lab ran a broader "distillation campaign" against its models, rather than a single isolated incident, according to The Information. The report doesn't yet detail what new evidence OpenAI is citing beyond the "campaign" framing itself.
What's New
Pulse previously covered OpenAI's original accusation that Moonshot's Kimi K3 model was trained in part by distilling outputs from Claude and OpenAI's own models -- a claim Moonshot denied at the time. The Information's new reporting marks an escalation in scope: OpenAI is now characterizing the behavior as an ongoing campaign rather than a one-off model, raising the stakes for how US labs and regulators treat open-weight Chinese models going forward.
“DeepSeek faced nearly identical allegations after its earlier releases.”
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 the broader accusation, and it previously denied the original single-model claim. 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.
