HighV Technology Geopolitics & AI8 September 2026, Tuesday
US intelligence agencies accuse six Chinese AI companies of distilling American models on an industrial scale
A joint advisory from the NSA, CISA and FBI targeted DeepSeek, Moonshot AI, Alibaba, MiniMax, StepFun and Z.AI; Beijing rejected the accusations.
According to Nextgov and The Register, the NSA, CISA and FBI issued a joint advisory on 8 September accusing six Chinese AI companies of training their own models since 2024 by extracting billions of tokens from US companies' models through fake accounts, bulk subscriptions and intermediary services. The advisory recommended that US AI companies covertly degrade the responses given to suspicious accounts.
According to The Next Web, Chinese Foreign Ministry spokesperson Mao Ning rejected the accusations, while the Commerce Ministry warned of a firm response. The advisory came two weeks before the Trump–Xi summit expected to be held in Washington on 24 September, turning AI models themselves into a new front of technology competition alongside chip export controls.
Talay assessment
Bottom line
The joint NSA, CISA and FBI advisory defines the outputs of AI models as a national security issue and shifts technology rivalry from chip export controls to the models themselves. In the near term the most likely outcome is US companies tightening account verification and access controls; formal sanctions steps depend on how the Trump–Xi summit expected on 24 September unfolds. The advisory adds a new item to the negotiating table rather than derailing the summit.
Likely effects
- Access to AI servicesNegative1–6 months
Targeting fake accounts, bulk subscriptions and intermediary services may lead US model providers to increase identity verification and regional access restrictions, which also makes access harder for legitimate users in third countries.
- US–China relationsNegativeWeeks
Beijing's denial and the Commerce Ministry's warning of a firm response heighten tension on technology ahead of the summit and raise the risk of reciprocal trade measures.
- Türkiye technology ecosystemUncertain1–6 months
Turkish start-ups and developers reaching US models through intermediary services may face stricter verification and possible cut-offs; US–China technology decoupling turns the choice of supplier into a political decision.
Possibilities, ranked
- 1Technical measures, summit on track60%
US companies tighten access controls and Washington defers formal sanctions; the summit takes place but the AI issue remains unresolved.
Watch: Access-policy announcements from US model providers and the 24 September summit going ahead
- 2Formal sanctions and Chinese retaliation25%
An export-control listing or similar restriction follows for the named companies; China responds with countermeasures and the technology front widens.
Watch: Named companies on US Commerce Department lists and countermeasure announcements from China's Commerce Ministry
- 3Technology tension eased at the summit15%
The leaders set up a dialogue mechanism on technology; the practical consequences of the advisory remain limited.
Watch: Joint language on AI or technology dialogue in the summit statement
Probabilities are calibrated judgement based on the sources, not measurement, and are revised as new information arrives. Not investment advice.