This composite source summary covers four primary and near-primary materials that together characterize the US-China AI national-security axis as of early 2026: two Center for Security and Emerging Technology (CSET) reports on Chinese military AI procurement, a US Department of Justice (DOJ) economic-espionage conviction, and Center for Strategic and International Studies (CSIS) and Department of Defense (DoD) materials on US military AI adoption. The four are filed together because the picture emerges from reading the procurement-data view (CSET), the enforcement view (DOJ/Ding), and the US-DoD-adoption view (CSIS/CDAO) against each other rather than from any single piece.
Sources covered
- CSET — China's Military AI Wish List (Probasco, Bresnick, McFaul, Feb 2026). Analysis of thousands of People's Liberation Army (PLA) open-source requests for proposals (RFPs), 2023–2024.
- CSET — Pulling Back the Curtain on China's Military-Civil Fusion (McFaul, Bresnick, Chou, Sep 2025). Analysis of 2,857 AI-related defense contract award notices covering 1,560 unique entities.
- DOJ press release — Former Google Engineer Found Guilty of Economic Espionage, Jan 30 2026 (Linwei Ding verdict).
- CSIS — Scaling AI-Enabled Capabilities at the DOD (Mar 2024), together with AFCEA/DefenseScoop reporting on the Task Force Lima sunset and the AI Rapid Capabilities Cell launch (Dec 11 2024).
Chinese military AI procurement (CSET)
The two CSET reports characterize Chinese military AI activity from open-source procurement records. The PLA is described as running thousands of short-cycle AI RFPs on 3–6 month timelines, with modest budgets, oriented toward C5ISRT (command, control, communications, computers, cyber, intelligence, surveillance, reconnaissance, and targeting). Stated priorities include AI decision-support drawing on open-source data; maritime and space capabilities framed as countering the US; biometric surveillance covering both face and gait recognition; and deepfakes for both offensive use and detection.
On the vendor base, the reports find that roughly 75% of vendors are nontraditional, reporting no state ownership, mostly founded after 2010, and framed as dual-use. These nontraditional vendors collectively win more contracts than the legacy state-owned enterprises (SOEs), though the SOEs — CETC, CASC, and NORINCO — still dominate the top-dollar awards. CSET presents this as military-civil fusion observable in procurement data rather than as doctrine alone, supported by the September 2025 report's dataset of 2,857 AI-related defense contract award notices across 1,560 unique entities.
US economic-espionage enforcement (DOJ / Ding)
On January 30, 2026, the DOJ announced what it described as the first US conviction on AI-related economic-espionage charges. Linwei Ding, a Google software engineer, was found guilty of stealing more than 2,000 pages of material on TPU architecture, GPU systems, SmartNIC, and cluster-communication software between May 2022 and April 2023. According to the DOJ, during the period of the theft Ding was negotiating to become chief technology officer of one PRC AI company and was founding another; public filings identify these as Rongshu Lianzhi and Shanghai Zhisuan. He was charged with seven counts of economic espionage under 18 U.S.C. §1831 (15-year maximum each) and seven counts of trade-secret theft under §1832 (10-year maximum each).
US military AI adoption (CSIS / CDAO / Task Force Lima)
Task Force Lima operated from August 2023 to December 2024, stood up to accelerate generative-AI adoption across the DoD. It catalogued 15 use-case areas spanning warfighting and enterprise functions and reviewed hundreds of workflows. On December 11, 2024, Task Force Lima was sunset and its mandate folded into the AI Rapid Capabilities Cell (AI RCC) under the Chief Digital and Artificial Intelligence Office (CDAO), with $100M in initial FY24/25 funding and $35M earmarked for four frontier-AI pilots. Radha Plumb led the CDAO at the time of the report, from April 2024 to January 2025; the post passed to Cameron Stanley in January 2026 after an acting interval. In December 2024 the office awarded roughly $40M in SBIR contracts to non-traditional vendors.
The operating model pairs CDAO with the Defense Innovation Unit (DIU): CDAO drives, while DIU pulls in commercial solutions. A minimum-viable capability for Combined Joint All-Domain Command and Control (CJADC2) was declared achieved through the GIDE V–VIII (Global Information Dominance Experiments) series.
CSIS panelists argued that the DoD should adopt a software-acquisition model built on persistent fielding and continuous updates, abandoning 12–18 month exercise cycles in favor of 90-day GIDE increments, and that it should focus not on building competing foundation models but on fine-tuning, orchestration, and containerized classified enclaves, described in the panel through a "the wild / the zoo / the cages" framing. One CDAO officer on the panel asserted that the DoD is "a week to two weeks behind" frontier commercial AI.
Key claims with confidence levels
| Claim | Confidence | Supporting sources |
|---|---|---|
| China's PLA actively procures AI for C5ISRT via thousands of short-cycle RFPs | High | CSET Feb 2026, CSET Sep 2025 |
| Military-civil fusion is operational in AI procurement (not rhetorical) | High | CSET Sep 2025 (2,857 contracts, 1,560 entities) |
| Ding is the first AI-economic-espionage conviction under §1831 | High | DOJ, FDD, CNBC |
| The stolen material constitutes the full stack for a hyperscale AI training system | Medium | DOJ description; inferred from TPU + GPU + SmartNIC + cluster-software combination |
| AI RCC / CDAO is the US institutional response to the PLA's speed advantage | Medium | CSIS, AFCEA — an argument rather than an established outcome |
| DoD is "a week to two weeks behind" frontier commercial AI | Low | Single CSIS panel assertion by a CDAO officer; self-reported |
Provenance
The four materials are additive rather than contradictory. The "different races" framing — under which the US optimizes for the frontier and China optimizes for deployment — is complicated but not contradicted: in the military lane, China appears to run a deployment-style playbook (short cycles, many vendors, civil-military fusion) while the US is attempting to build one (AI RCC, 90-day GIDE).
Relationships
- related: AI and National Security
- related: Export Controls (AI) — industrial espionage (Ding) is the insider-threat counterpart to chip export controls, both enforcement arms of the same strategy, adding a dimension beyond Amodei/Ord/Ball.
- related: AI Race Dynamics — adds the military and national-security dimension to what had been framed primarily as commercial-frontier and safety-capability races.
- related: Clawed — Ball's critique of the Anthropic-DoW conflict sits in the same ecosystem as the AI RCC and CDAO posture.
- related: Autonomous Weapons — the C5ISRT focus is the sensor and decision-support side of the same military-AI question that DoD Directive 3000.09 governs on the kinetic side.
- related: US-China AI Competition: Different Races, Different Metrics — adds the military lane alongside the compute, capability, and deployment lanes.