CISA advisory names Moonshot AI among Chinese firms distilling US AI models
- Place in VP50
- #20
- Source
- 1
- Claims verified
- 5 of 5
What happened
On 8 September 2026 the US Cybersecurity and Infrastructure Security Agency published advisory AA26-251A, a joint product of the NSA, CISA and the FBI, on China-based AI companies it said were running large-scale distillation campaigns against US frontier AI companies. The advisory names six companies: DeepSeek, Moonshot AI, Alibaba, MiniMax, StepFun and Z.AI. It says they extracted billions of tokens across millions of exchanges or requests from US frontier models, likely with Chinese government awareness.
On Moonshot AI, the advisory states that the company has conducted a widespread distillation campaign against US frontier AI companies since at least mid-2025. It says Moonshot AI used Claude Fable 5 data to train its Kimi-K3 model and GPT-4o data to train Kimi-K2. The advisory gives no token count specific to Moonshot. It describes methods such as fraudulent or shared accounts and gray-market proxies. It sets out detection and information-sharing steps for US AI companies and does not describe any penalty, sanction or enforcement action against Moonshot or the other firms.
Key points
- CISA, NSA and FBI advisory AA26-251A, dated 8 September 2026, names Moonshot AI among six China-based firms.1
- The advisory says Moonshot AI has run a distillation campaign against US frontier AI companies since at least mid-2025.1
- It says Moonshot AI used Claude Fable 5 data to train Kimi-K3 and GPT-4o data to train Kimi-K2.1
- No penalty, sanction or enforcement action is described; the advisory is defensive guidance.1
Why it matters
A US agency advisory names Moonshot AI and alleges that distillation of US models was used to train its Kimi models. This is an allegation by authorities, not a court finding.
The advisory describes no penalty or enforcement step. It adds a public record from the US government naming the company, which bears on the regulatory exposure that a valuation rests on.
The advisory says distillation forms the core of these firms' AI development strategy. For Moonshot that links the Kimi models to access to US-built systems, as the advisory describes it.
In context
| Round | Date | Post-money valuation | Raised | Led by |
|---|---|---|---|---|
| Angel round | Jun 13, 2023 | $300M | $200M | Unknown |
| Series A+ | Feb 20, 2024 | $2.5B | $1.0B | Alibaba |
| Series B | Aug 5, 2024 | $3.3B | $300M | Unknown |
| Series C | Dec 2025 | $4.3B | $500M | IDG Capital |
| Funding round | Feb 2026 | $10B | $700M | 九安医疗 (Jiuan Medical) and 五源资本 (5Y Capital) |
| Series F | Jul 29, 2026 | $35B | $3.5B | National AI Industry Investment Fund |
Updated
Fact-check
- CISA published advisory AA26-251A on 8 September 2026, jointly with the NSA and FBI.Confirmed by [1] CISA: the advisory is dated 8 September 2026 and is a joint product of the NSA, CISA and FBI.Verified
- The advisory names six companies: DeepSeek, Moonshot AI, Alibaba, MiniMax, StepFun and Z.AI.Confirmed by [1] CISA: the advisory lists these six companies and describes each one's alleged campaign.Verified
- The six extracted billions of tokens across millions of exchanges, likely with Chinese government awareness.Confirmed by [1] CISA, which uses this wording for the six companies together.Verified
- The advisory says Moonshot AI has run a distillation campaign since at least mid-2025, training Kimi-K3 on Claude Fable 5 data and Kimi-K2 on GPT-4o data.Confirmed by [1] CISA: the Moonshot AI section states the mid-2025 start and names Claude Fable 5 data for Kimi-K3 and GPT-4o data for Kimi-K2.Verified
- The advisory describes no penalty or enforcement action.Confirmed by [1] CISA: the document is defensive guidance on detection, response and information sharing, with no sanctions or enforcement described.Verified
Sources
- [1]CISA (opens in a new tab)China-Based Artificial Intelligence Companies Conducting Industrial-Scale Distillation Campaigns Against U.S. AI CompaniesMoonshot AI has conducted a widespread distillation campaign against U.S. frontier AI companies since at least mid-2025.
