Research: The China Parallel — the Body-Control Stack

July 2, 2026

The China parallel to the US/UK/Israel biometric axis, held to the same sourcing discipline as the rest of this corpus: peer-reviewed papers, government documents (Chinese ministry releases, OHCHR, US BIS/Commerce, FCC), company filings (Shenzhen Stock Exchange, Hong Kong Stock Exchange), arXiv technical papers, named investigative outlets with documented methodology (ASPI, Citizen Lab, IPVM, HRW with reverse-engineered forensic artifacts on GitHub, BuzzFeed News satellite analysis), and established trade press only after that. No anti-CCP fabrication, no pro-CCP apologia. Where a topic is contested or sourced from a single advocate, both the claim AND the documented limit are surfaced.

Dates given as ISO calendar. Every claim about an identifiable individual or named corporation carries a source URL. Where a characterization is a judgment made by a named outlet or analyst, it is attributed as such rather than adopted as our own voice.

A note on the threat model for sourcing on China: the topic attracts both pro-CCP apologetics (“nothing to see here, this is just credit scoring”) and anti-CCP fabrication (“a billion people walk around with a single Black Mirror score”). The discipline below is to cite primary documents (Chinese government white papers in translation via China Law Translate, Shenzhen Stock Exchange disclosures, OHCHR), peer-reviewed empirical work (Kostka, Daum, ASPI Xinjiang Data Project, Citizen Lab reverse-engineering), and named investigative work with reproducible methodology (BuzzFeed satellite + architectural analysis, HRW IJOP app reverse-engineering with forensic artifacts published on GitHub). Adrian Zenz is cited because his work is documentary-driven (publicly available Chinese government records, leaked Xinjiang Police Files) but his Victims of Communism Memorial Foundation affiliation is noted because methodology critics have raised it; his core documentary findings have been independently corroborated by ASPI, BuzzFeed satellite work, and OHCHR.

1. Social credit — the fragmented reality vs. the unified-score myth

The single most-mistranslated topic in Western coverage of China. The record here is structured around four propositions, each documented.

Proposition 1 — There is no unified citizen score. The phrase “social credit code” in the official 2014-2020 planning outline (国务院关于印发社会信用体系建设规划纲要的通知, State Council Notice 2014/No. 21) refers to a unified social credit identification number for legal persons (mostly companies) — analogous to a US EIN — not a personal scoring rank. Jeremy Daum (Senior Fellow, Yale Law School Paul Tsai China Center; founder of China Law Translate) has documented this in depth, including direct translation of the planning documents. The original 2014-2020 social credit planning outline did not mention numerical citizen scores at all. Sources: Jeremy Daum, “China through a glass, darkly” (China Law Translate); Daum on “Getting Rongcheng Right”; Daum, Yale Paul Tsai China Center page; The China Project interview, 2022-02-03.

Proposition 2 — There are real city-level pilots with numerical scores, but they are local, voluntary in practice, and fragmented.

Proposition 3 — The commercial pilot (Sesame Credit / Zhima Credit) was denied a permanent license by the People’s Bank of China. In 2015 the PBOC issued provisional zhengxin (personal credit reporting) licenses to eight private firms — Ant Group’s Sesame Credit, Tencent Credit, plus six others — for a two-year pilot. In June 2017 the PBOC refused to extend any of the eight. Citing conflict-of-interest (the firms’ priorities lay in their own e-commerce / payment ecosystems) and inability to assign comprehensive independent credit scores, the PBOC instead issued a single license in 2018 to Baihang Credit (百行征信), jointly owned by the National Internet Finance Association of China (36%) and the eight commercial firms (each 8%). Sesame Credit became a loyalty / merchant-discount program, not a state credit instrument. This is the load-bearing fact that the “China has a unified Black Mirror score” thesis cannot survive. Sources: Zhima Credit — Wikipedia with PBOC license history; Caixin Global on Tencent launching a Sesame competitor and the licensing landscape; Genia Kostka, “China’s social credit systems and public opinion,” New Media & Society (2019).

Proposition 4 — What is real and consolidated is the corporate-side social credit system: a unified blocklist / redlist regime for legal persons. (Program mechanics are documented; the “coercive” reading belongs to the cited analysts.) The blocklist (失信被执行人名单, List of Persons Subject to Enforcement for Breach of Trust) is maintained by the Supreme People’s Court and integrated with the National Credit Information Sharing Platform (全国信用信息共享平台). As of late 2024 the blocklist contained millions of individuals and entities barred from purchasing flights, high-speed rail tickets above a class threshold, and certain real-estate purchases — but blocklist entry is triggered by court judgment of refusal to perform obligations, not by an algorithmic score. A 2025 update to the Social Credit Code formalised cross-ministry data sharing so corporate violations in tax, labor, environmental, and customs columns block access across the others. Sources: Supreme People’s Court blocklist portal (Chinese); Daum, “Social Credit Action in 2025”; Daum, “Credit Risk Classification: A social credit score”.

The empirical-survey finding — Kostka’s New Media & Society survey (n ≈ 2,209 respondents across Chinese regions, 2018) found roughly 80% of respondents either “somewhat approve” or “strongly approve” of social credit systems, with wealthier / better-educated / urban / older respondents most supportive. They framed it as enforcing honesty in commerce, not as an instrument of repression. This is the inverse of the Western media frame and is the empirical finding to engage if the chapter wants to do anything more sophisticated than retail outrage. Source: Kostka 2019, New Media & Society; MERICS, “What do young Chinese think about social credit? It’s complicated”.

2. Skynet and Sharp Eyes — the actual CCTV state

Skynet (天网, Tianwang). Launched 2005. Urban CCTV + facial-recognition program led by the Ministry of Public Security. As of 2019 Chinese state media (Global Times) and IPVM analyses reported approximately 200 million Skynet cameras in mainland China; by 2024 estimates (including private + public combined) reached the 500-600 million range across all surveillance cameras (camera counts are estimates from IPVM / state media, not audited totals; see Section 13). Note: that combined figure includes Sharp Eyes rural cameras, commercial cameras pulled into the police data-fusion platforms, and private residential cameras — the headline “600 million Skynet cameras” framing conflates programs. Source: IPVM, China Public Video Surveillance Guide: From Skynet to Sharp Eyes; Wikipedia, Mass surveillance in China (citing Chinese state media); SCMP on Skynet.

Sharp Eyes (雪亮工程, Xueliang Gongcheng — literally “Bright Snow Project”). Launched 2015, codified in the 2016 13th Five-Year Plan with the goal of 100% video coverage of China’s “public spaces and key industries” by 2020. Sharp Eyes extends Skynet into rural areas and ingests private CCTV (residential, commercial, transport-hub) into the same data-fusion platforms. CSET’s analysis (Dahlia Peterson, Center for Security and Emerging Technology, Georgetown) documents the program in detail and notes that as of available reporting the country has come “very close” to the 100% goal without confirming it. Sources: CSET — “China’s ‘Sharp Eyes’ Program Aims to Surveil 100% of Public Space”; CSET — “How China harnesses data fusion to make sense of surveillance data”; Brookings analysis.

Data fusion (“societal resource integration platforms”). (Documented by CSET/Georgetown.) The point of Sharp Eyes is not the cameras themselves but the unified data fusion at the back end. Per CSET, surveillance data is combined with GIS data and sent to “societal resource integration platforms” (社会资源整合平台) operating in Xinjiang and at least four other provinces, which combine: facial recognition, vehicle and license plate recognition, MAC addresses, mobile phone IMEI numbers, WeChat account IDs, hospital records, hotel registrations, internet cafe logs. This is the structural innovation — not the camera count. Source: CSET data fusion piece; CSET, Designing Alternatives to China’s Repressive Surveillance State (October 2020 PDF).

Accuracy claims. Chinese state media (Global Times) has claimed 99.8% facial-recognition accuracy and the ability to scan all 1.4 billion mainland residents in under a second. These are state-media claims, not independently audited. NIST’s Face Recognition Vendor Test (FRVT) shows Chinese vendors (SenseTime, Megvii, Yitu) among the top performers on the NIST 1:N identification benchmark, supporting the capability claim while leaving the deployment-scale claim open. Source: NIST FRVT 1:N Identification leaderboard.

3. The “AI surveillance majors” — corporate filings, sanctions, and product capabilities

The five firms that anchor the Chinese state’s AI-surveillance stack. All five were added to the BIS Entity List on 2019-10-09 (Federal Register, “Addition of Certain Entities to the Entity List,” 84 FR 54002) — the stated reason, per the US government’s finding, being involvement in human-rights violations against Uyghurs in Xinjiang. The Entity List addition restricted them from receiving items subject to the EAR without a BIS license. Source: Federal Register, 2019-10-09 addition; BIS press release archive; CSIS — Understanding the Entities Listing in the Context of U.S.-China AI Competition.

Hikvision (Hangzhou Hikvision Digital Technology Co., Ltd.) — ticker 002415.SZ (Shenzhen). Listed on Shenzhen Stock Exchange since 2010 (not Hong Kong). Largest shareholder is China Electronics Technology Group Corporation (CETC) via subsidiary, a state-owned military-electronics conglomerate.

Dahua Technology (Zhejiang Dahua Technology Co., Ltd.) — ticker 002236.SZ (Shenzhen). Listed Shenzhen 2008. ~10.5% global surveillance-camera market share 2024. IPVM and LA Times jointly documented in late 2020 that Dahua had developed and shipped facial-recognition software with an explicit “Uyghur alert” filter — code that flagged subjects detected as ethnically Uyghur (attributed to the IPVM / LA Times reporting). Source: Dahua — Wikipedia; IPVM report (referenced in Mass surveillance in China); Federal Register Entity List addition.

SenseTime Group Inc. — 0020.HK (Hong Kong Stock Exchange). Listed Hong Kong 2021-12-30 after a US-sanction-driven delay (the original IPO was halted December 2021 when the US Treasury added SenseTime to the Non-SDN Chinese Military-Industrial Complex Companies List, prohibiting US investor purchases). Headquartered Hong Kong; partly state-owned. Founded 2014, originally from a Chinese University of Hong Kong lab.

Megvii Technology (旷视科技, Beijing Megvii Co., Ltd.) — operator of Face++. Three failed IPO attempts 2019-2024.

iFlytek (科大讯飞 — Anhui USTC iFlytek Co., Ltd.) — ticker 002230.SZ (Shenzhen). Voice-recognition specialist. Per HRW’s 2017 reporting, iFlytek is the designated supplier of voice-pattern collection systems purchased by Xinjiang and Anhui police bureaus. iFlytek’s own website states it has helped the Ministry of Public Security build a national voice-pattern database and that it has co-founded, with the MPS Forensics Center, a key ministry laboratory in AI voice technology. Headquartered in Anhui Hefei High-tech Industry Development Zone — one of at least 28 MIIT-designated national-level military-civilian fusion bases. BIS Entity List 2019-10-09. MIT terminated its iFlytek research agreement 2020-04 citing human-rights concerns. Sources: Human Rights Watch — “China: Voice Biometric Collection Threatens Privacy” (2017-10-22); iFlytek — Wikipedia, with MPS and MIIT cites; Washington Times on MIT termination.

4. Xinjiang — the Uyghur surveillance case

The OHCHR August 2022 report — the load-bearing UN document. (That the report exists and says what is quoted is documented; its conclusions are attributed to OHCHR.)

On 2022-08-31, the Office of the United Nations High Commissioner for Human Rights (then-Acting High Commissioner Nada Al-Nashif, signed by outgoing High Commissioner Michelle Bachelet 13 minutes before the end of her tenure) released the “Assessment of human rights concerns in the Xinjiang Uyghur Autonomous Region, People’s Republic of China.” Key findings — quoted-language matters here:

Methodology: based on Chinese government documents and interviews with victims and survivors. The Government of China issued a 130-page response, included by OHCHR as an annex. Sources (primary): OHCHR Assessment PDF (full 2022-08-31 final assessment); OHCHR press release announcing the assessment; OHCHR country-reports page; UN News story. Subsequent: Amnesty International, 3rd anniversary statement, 2025-08.

ASPI Xinjiang Data Project. The Australian Strategic Policy Institute (Canberra) compiled the most systematic open-source documentary record (attributed to ASPI). Includes:

The Xinjiang Police Files (2022). (Documentary leak; publication + contents attributed to the 14-outlet consortium and to Zenz, whose affiliation is flagged below.) Approximately 10 GB of leaked Xinjiang public-security files — speeches by Chen Quanguo, internal protocols, more than 2,884 photographs of detainees, security spreadsheets. Forwarded to anthropologist Adrian Zenz from an anonymous source. Published 2022-05-24 in coordinated release by an international consortium of 14 media organizations. Files cover Konasheher (Shufu) and Tekes counties; reference internment of 20,000+ individuals.

BuzzFeed News satellite analysis (Pulitzer 2021). (Attributed to the BuzzFeed News team.) Megha Rajagopalan (reporting, London), Alison Killing (architect, Netherlands), Christo Buschek (programmer, Berlin) used satellite imagery (Planet Labs tasking, plus comparison of censored Baidu Maps tiles vs. uncensored international tiles to locate camps via the censorship signature itself) and architectural analysis to identify and analyze more than 260 facilities bearing the hallmarks of fortified detention compounds, built since 2017. Subsequent capacity analysis of 347 compounds against Chinese prison-construction standards estimated capacity to hold “more than 1 million” simultaneously. Awarded the 2021 Pulitzer Prize for International Reporting.

IJOP — Integrated Joint Operations Platform (一体化联合作战平台). (Attributed to HRW; the app decompilation + forensic artifacts are reproducible from the GitHub repo.) Predictive-policing data-fusion system documented through HRW’s reverse-engineering of its associated police mobile app. Between January 2018 and February 2019, HRW (lead researcher Maya Wang) decompiled and analyzed the IJOP-affiliated police app used in Xinjiang. The reverse-engineering produced a 68-page report and a forensic-artifact repository published on GitHub.

Biometric collection scale. (Attributed to HRW.) Per HRW’s 2017 voice-biometric reporting and 2017 China’s Bloody Harvest / Pixelated Privacy reporting, Xinjiang authorities mandated DNA, fingerprint, iris scan, and voice-pattern collection from all residents aged 12-65 in the “Physicals for All” (全民健康体检工程, Quan Min Jian Kang Ti Jian Gong Cheng) campaign starting 2016-2017. Sources: HRW — “China: Voice Biometric Collection Threatens Privacy” (2017-10-22); HRW — China: Minority Region Collects DNA from Millions.

5. e-CNY (digital yuan) + biometric / ID architecture

Adoption data, end-November 2025 (PBOC release). Cumulative e-CNY transactions: 3.4+ billion / RMB 16.7 trillion (~USD 2.3 trillion). +800% on 2023. 17 provinces piloting. ~225 million personal digital wallets opened. Effective 2026-01-01, a new e-CNY measurement and management framework took effect, reframing e-CNY less as “digital cash” and more as deeper integration with the regulated financial system (including deposit features rolling out 2026). Sources: gov.cn English on the 2026 framework; Atlantic Council — “What to watch as China prepares its digital yuan for prime time”; Atlantic Council CBDC Tracker — China.

Identity tie-in. e-CNY wallets are tiered by KYC level. Tier 1 (anonymous-ish, low-value, ~RMB 2,000 daily cap) requires phone number only; the wallet provider can still match the phone number to a real identity via the State Council mobile-number real-name registration system (实名制). Tier 4 (full-feature, high-value) requires bank account linkage and resident-ID card binding (居民身份证), which is China’s second-generation biometric identity card carrying name, photo, fingerprint chip, and the 18-digit national ID number. Source: CNAS — China’s Digital Currency: Adding Financial Data to Digital Authoritarianism (Yaya J. Fanusie and Emily Jin, January 2021). CNAS’s characterization (attributed to CNAS): “the DCEP technical architecture will allow for real-time or near-real-time financial surveillance of all users’ transactions.”

Facial-recognition checkout integration. Both Alipay (Ant Group / Alibaba) and WeChat Pay (Tencent) launched facial-recognition checkout systems (“Smile to Pay” / Dragonfly devices for Alipay; Frog devices for WeChat Pay) starting 2018-2019. Deployed in quick-service restaurants, supermarkets, vending kiosks, and Beijing / Shanghai / Guangzhou metro systems. Biometric templates are stored on encrypted servers per PRC cybersecurity law; the facial template is linked to passport (for foreigners) or resident-ID (for nationals). For high-value transactions the biometric is the security gate.

Project mBridge (multi-CBDC bridge). BIS Innovation Hub project, with PBOC (e-CNY), Hong Kong Monetary Authority (e-HKD), Bank of Thailand, Central Bank of UAE, and (since 2024) the Saudi Central Bank as full participants. The BIS exited the project in late 2024 after concerns that mBridge could be used to evade sanctions; governance shifted fully to the participating central banks. As of late 2025 mBridge cross-border transaction volume had grown ~2,500× from early-2022 pilots to approximately USD 55.49 billion, with e-CNY making up >95% of settlement volume on the platform. Cross-border retail e-CNY testing is live in Hong Kong, Macau, Laos, Thailand, Cambodia, and Singapore. Sources: BIS Project mBridge page; Ledger Insights on the $2 trillion / mBridge scale; Fintech News HK; Tandfonline academic article on mBridge state-finance-tech nexus.

6. Chinese frontier AI labs — the DeepSeek moment and after

The structural fact: the US export-control regime explicitly assumed that compute-rationing would keep the Chinese frontier ~12-18 months behind. DeepSeek-R1 collapsed that assumption in a week.

DeepSeek (深度求索) — DeepSeek-V3 (December 2024) and DeepSeek-R1 (2025-01-20). Founded mid-2023 by Liang Wenfeng (梁文锋), built on his hedge fund High-Flyer Quant’s GPU infrastructure (~10,000 NVIDIA A100s acquired pre-export-control + ~2,048 H800s acquired afterward). R1 released MIT license on 2025-01-20 with accompanying arXiv paper. Within a week the DeepSeek mobile app topped the Apple US App Store free chart, the first Chinese AI product to beat ChatGPT in US consumer download rankings.

Moonshot AI (月之暗面) — Kimi. (Founding + release documented; valuation attributed to Silicon Republic.) Founded 2023 by Yang Zhilin. Kimi K1.5 release in early 2025 matched R1 on several reasoning benchmarks; closed-weights, so adoption was limited. Backed by Alibaba and Tencent. Per Silicon Republic: USD 2 billion raise in early 2026 at USD 20 billion valuation. USD 1.4 billion in cash reserves reported. Source: Silicon Republic — “Moonshot AI valued at $20bn after $2bn raise for Kimi creator”; Recode China AI on DeepSeek and Moonshot.

Zhipu AI (智谱) — GLM / ChatGLM. (Listing + raise documented; “state-adjacent” is an analytical framing.) State-adjacent. Strategic investment from Saudi Arabia’s Prosperity7 sovereign-tech fund. Listed Hong Kong Stock Exchange January 2026 alongside MiniMax — combined raise over USD 1.1 billion; Zhipu market cap ~USD 13 billion post-listing. Source: Recode China AI on the Hong Kong listings; France 24 — China’s top AI players (April 2026).

Alibaba — Qwen team. Qwen 2.5 family open-weights releases continuing through 2025-2026; Qwen3-Max claimed at the closed-source frontier. Xinhua reported 2026-02 that a coordinated “next DeepSeek moment” race was underway with multiple labs releasing rapid open-source successors. Source: Xinhua, 2026-02-12; Xinhua, 2025-07-31 — China’s AI firms roll out DeepSeek rivals.

Baidu — Ernie. Ernie 4.0 / Ernie 4.5 closed-weights. Baidu pivoted to releasing Ernie 4.5 open-weights mid-2025 under competitive pressure.

01.AI (零一万物). Founded by Kai-Fu Lee (former Google China head). Yi family of open-weights models. Pivoted from frontier ambition toward enterprise inference services after DeepSeek’s pricing collapse made consumer chatbots unprofitable.

Tencent — Hunyuan. Internal lab; closed-weights primarily.

ByteDance — Doubao. Consumer-facing chatbot, leveraged TikTok / Douyin distribution.

Step Stones / StepFun. (Attributed to Hello China Tech.) Backed Yin Qi (former Megvii founder, after three failed Megvii IPOs); USD 720M strategic distribution bet reported 2025. Source: Hello China Tech on StepFun.

7. IJOP and predictive-policing — see Section 4

IJOP is the canonical case and lives under Xinjiang. No separate national-level predictive-policing equivalent is openly documented at the same level of forensic detail; Sharp Eyes data fusion is the closest non-Xinjiang analog.

8. DJI and the drone state

DJI (SZ DJI Technology Co., Ltd., 深圳大疆创新). Privately held, headquartered Shenzhen. ~70% global drone market share; ~77% US market share; ~80% US consumer drone market. Roughly half of all drones sold in the US are DJI-made.

Note: direct People’s Armed Police (PAP) procurement of DJI hardware is not documented to primary-source standard (see Section 13).

9. Export of Chinese surveillance tech — the AI surveillance footprint

The Carnegie AIGS Index (Feldstein 2019, with updates). (Attributed to Feldstein / Carnegie.) Steven Feldstein’s AI Global Surveillance Index documented AI surveillance tech use in 176 countries. Chinese companies — particularly Huawei, Hikvision, Dahua, and ZTE — supplied AI surveillance technology to 63 countries, 36 of which were Belt-and-Road signatories. Huawei alone was responsible for AI-surveillance deployments in at least 50 countries. The next largest non-Chinese supplier was Japan’s NEC Corporation (14 countries). Feldstein’s blunt finding: “China is exporting surveillance tech to liberal democracies as much as it is targeting authoritarian markets.” Sources: Carnegie — The Global Expansion of AI Surveillance (Feldstein 2019); PDF of the full working paper; AIGS Index PDF.

Huawei Safe City deployments — named contracts.

The African Union Headquarters incident (2012-2017). (Attributed to Le Monde Afrique 2018 and Reuters 2020; Huawei and China denied it.) The USD 200 million AU headquarters in Addis Ababa was built and paid for by the Chinese government and opened January 2012. Computer systems including servers, routers, and PBX were predominantly supplied by Huawei and ZTE. Per Le Monde Afrique (2018-01), confidential AU data was being exfiltrated from the headquarters IT network to Shanghai servers nightly between midnight and 2 a.m. for five years (2012-2017), until discovered by an AU technical staff member in January 2017. The AU kept the discovery secret for a year. Huawei and the Chinese government denied the allegations. Subsequent reporting by Reuters (2020) and Privacy International documented continued anomalous data exfiltration involving the AU’s CCTV system supplied by Hikvision. Sources: New America Foundation — What’s the Deal with Huawei and This African Union Headquarters Hack?; Wikipedia — 2018 China-African Union espionage allegations; Privacy International — China spies on African Union Addis Ababa; CFR analysis; International cyber law toolkit (CCDCOE).

Southeast Asia / Middle East / Latin America. (Attributed to Feldstein and CSIS.) Deployments documented in Malaysia (Kuala Lumpur Safe City), Pakistan (Islamabad Safe City), Saudi Arabia (NEOM smart-city contract via Huawei), UAE, Ecuador (ECU-911 emergency-response system with Huawei + CEIEC equipment — extensively documented by NYT in 2019), Venezuela, Bolivia. Source: Feldstein, AI Global Surveillance Index.

10. Western tech inside the Chinese stack — the gap between stated and actual

The defining hypocrisy of the US export-control regime: the controls are real on paper, partially effective in slowing Huawei Ascend / Cambricon chip ramp, and substantially leaky on the consumer / enterprise GPU flow.

NVIDIA chip smuggling — documented operations.

Trump-era reversal (July 2025). The Trump administration lifted the H20 / China-AI-chip ban, clearing NVIDIA and AMD to resume legal sales of restricted variants to China. Per Built In and The Register, the policy reversal substantially reopened the legal channel that smuggling had been routing around. Sources: Built In — Trump Lifted the AI Chip Ban on China; The Register — The risks of export controls on AI chips.

Apple — iCloud China (Guizhou-Cloud Big Data). (The architecture / key migration is documented; the “state can compel decryption” reading is attributed to security researchers and NYT/Reuters.) Apple’s first China data center opened May 2021 in Guiyang, Guizhou Province, operated by Guizhou-Cloud Big Data Industry Development Co., Ltd. (GCBD) — a state-owned enterprise wholly owned by the Guizhou provincial government. From 2018 onward, Apple required Chinese iCloud customers to accept new terms naming GCBD as service provider and Apple as “an additional party.” The encryption keys to unlock iCloud data — previously stored exclusively in the US — were moved into China to satisfy PRC cybersecurity-law requirements. Per multiple independent security researchers and NYT / Reuters reporting, the practical effect is that the Chinese state can issue a domestic legal order to GCBD compelling decryption of data on Chinese users, an option not available under prior architecture. Sources: Data Centre Magazine — What is happening in Apple’s data centres in China?; The Register — Apple hands Chinese iCloud to Guizhou-Cloud Big Data; The Hacker News on Chinese government access; Amnesty International — When Profits Threaten Privacy; Bangkok Post — Censorship, surveillance and profits.

WeChat surveillance of international users — Citizen Lab. (Attributed to Citizen Lab, University of Toronto.) Citizen Lab (Munk School; Director Ronald Deibert) published 2020-05 “We Chat, They Watch” documenting that WeChat surveils content shared between non-China-registered accounts and uses the surveilled content to train the censorship classifiers applied to China-registered accounts. Image surveillance is by file-hash matching against a sensitive-hash index, performed in real time, with cache miss escalation to deeper content inspection. International users training the censorship apparatus for mainland users. Subsequent 2023 report (Privacy in the WeChat Ecosystem) and 2024 report (Security Analysis of WeChat’s MMTLS Encryption Protocol) extend the analysis. Sources: Citizen Lab — “WeChat Surveillance Explained” (2020-05); Citizen Lab — “We Chat, They Watch” full report; Citizen Lab — Privacy in the WeChat Ecosystem (2023-06); Citizen Lab — MMTLS encryption analysis (2024-10).

Intel / AMD / Microsoft / Oracle presence. (The presence is documented; the closing “Western-hardware-on-Chinese-control” reading is analytical framing.) Beyond the smuggling story, the legal channel still runs. Intel Xeon and AMD EPYC chips power most non-frontier Chinese enterprise datacenters; Microsoft Azure China is operated by 21Vianet under a license-arrangement structure (legally a separate company); Oracle and SAP enterprise software is licensed into PRC-resident operating companies. The point is not that the Chinese AI stack is autarchic — it isn’t — but that it is “Western-hardware-on-Chinese-control” much more than the export-control framing suggests.

11. Chapter-anchor incidents

A short ranked list of incidents that work as load-bearing chapter anchors. Each is sourced to a primary document. (These re-point to fully-documented claims above. Where an anchor rests on attributed reporting rather than a primary record — the AU exfiltration, the GPU-smuggling charges — the attribution and the charged/alleged framing carry through into the prose.)

  1. The PBOC denying Sesame Credit a license in June 2017. The single most counter-intuitive fact in the record. Western coverage assumes the Chinese state runs the consumer credit score; the actual record is that the state’s central bank stripped Alibaba of the option to build a unified consumer credit score, judged it would entrench Alibaba’s e-commerce monopoly, and replaced it with a state-supervised utility (Baihang) jointly owned by the same eight firms. The control move was from Beijing against Hangzhou — Alibaba HQ. Source: Zhima Credit Wikipedia with PBOC license history.

  2. The IJOP seven-day round-up: 15,683 Xinjiang residents detained in one week (June 2017). Documented in a leaked Xinjiang public-security document reverse-engineered through the IJOP police app by HRW. The operational scale of algorithmic detention reduced to a single number. Source: HRW China’s Algorithms of Repression.

  3. The African Union Headquarters data exfiltration (2012-2017). A USD 200M building gifted by China to the AU sent the AU’s internal data to Shanghai every night for five years, undetected, on Huawei equipment (attributed to Le Monde Afrique; Huawei and China denied it). The cleanest single anecdote for “infrastructure aid is surveillance acquisition.” Source: Le Monde Afrique via Wikipedia entry with cite trail.

  4. NVIDIA’s $589 billion one-day market-cap loss, 2025-01-27. The DeepSeek-R1 reveal — open-weights, MIT-licensed, training-cost claim of $5.9M total — collapsed the assumption that compute-rationing was a working containment strategy. Source: The Register on DeepSeek training cost and market impact.

  5. The Hao Global GPU smuggling case — $160M of H100s diverted via Taiwan and Southeast Asia in seven months (Oct 2024 - May 2025). (Charges, not convictions.) The empirical refutation of US export-control efficacy at the unit-level rather than the model-output level. Source: CNBC.

  6. Apple moving iCloud encryption keys to GCBD in Guizhou (2018-2021). The single neatest case of “Western tech inside the Chinese stack” — the US’s most valuable consumer-electronics company has explicitly engineered its Chinese product to be readable by the Chinese state. Source: Bangkok Post / NYT republication of the architectural compromise.

  7. Hikvision FY2024 revenue: USD 12.9 billion, 28% international, 23% global market share. The Entity-Listed firm growing through it. Source: Hikvision 2024 results page.

12. The comparison-with-US structural finding

The cleanest finding for the body-and-convergence argument, stated with the precision the topic requires. This section is an analytical equivalence argument — explicitly not “the US is becoming China.” The underlying program facts on the US side are sourced in the sibling body-layer references (see Related research); the symmetric standard holds — neither soften documented Chinese state behavior nor overstate the US comparison beyond what those references support.

The US and China have built the same control surface; they differ in who staffs the back office. Both states have CCTV at saturation density in urban cores (US TSA/CBP biometric exit ran facial comparison on 200+ million travelers in FY2024; China’s Skynet/Sharp Eyes claims approach 100% public-space coverage). Both run national biometric ID systems (US: REAL ID enforcement 2025-05-07, FBI NGI ~145M civil-fingerprint enrollments, DHS HART; China: second-generation Resident ID card with chip + national ID number). Both have moved consumer payment toward biometric-gated digital rails (Apple Pay + mDL + state mobile wallets; e-CNY + Alipay/WeChat facial checkout). Both have moved health and genetic data into private corporate vaults whose contents can be transferred without re-consent (US: 23andMe→TTAM, Ancestry, FamilyTreeDNA; China: National Genomic Data Center, BGI Group with reported tie to state biobanks). Both have deeply integrated predictive-policing platforms (US: Palantir / Geolitica / PredPol descendants; China: IJOP, Sharp Eyes data-fusion).

The structural difference is consolidation under a single political authority versus consolidation under a federation of corporate, federal-agency, and state-government authorities with overlapping jurisdictions. China’s system has one back office. The US system has thirty-six (call it: NSA, FBI, DHS, CBP, IRS, SSA, plus 50 state DMVs, plus the major cloud providers running it as compute, plus the data brokers, plus the credit bureaus). That is the comparison the chapter should make: the US has built something architecturally comparable to the Chinese stack, with the federation of operators producing a Madisonian dilution that may or may not survive a unified-purpose actor seizing the consolidated stack at speed.

The China parallel is therefore not “we are becoming China.” It is: the US has assembled the same control surface that China has assembled, on commercial-incentive logic, and is one consolidation event away from operationally equivalent capability. That is the load-bearing structural finding.

13. What is NOT verifiable

Discipline note: the following claims appear in the adjacent literature and have not been corroborated to primary-source standard. All are documented here as the myth-vs-fact ledger.

Verdict

Bottom line: The documented Chinese control-infrastructure record is strong enough to carry a chapter without any of the mistranslated tabloid myth-material — and is strongest precisely where the popular Western frame is weakest. The load-bearing facts are all primary or named-investigative with resolving sources: no unified citizen score (State Council Notice 2014/No.21; Daum; Kostka’s peer-reviewed survey), the PBOC’s 2017 refusal of Sesame Credit a license, the Skynet/Sharp Eyes data-fusion architecture (CSET/Georgetown), the five surveillance majors’ US-government listings (Fed. Reg. 84 FR 54002; FCC; DOD 1260H) and their exchange-filed financials, the OHCHR 2022 “may constitute crimes against humanity” assessment with China’s annexed rebuttal, HRW’s reverse-engineered IJOP forensics, BuzzFeed’s Pulitzer satellite work, the e-CNY/mBridge rails, the DeepSeek shock, and the export footprint via Feldstein’s AIGS Index.

The discipline that makes it publishable: every state-surveillance CHARACTERIZATION is attributed to the reporting or analyst that made it, while the documented PROGRAM facts stand as facts; criminal matters (GPU-smuggling prosecutions) are labeled as charges, not convictions; contested investigators (Zenz) are surfaced with their affiliation and corroboration chain rather than laundered past; and the myth-material is quarantined in Section 13 as unsupported. The symmetric standard holds in both directions — no pro-CCP defanging of the OHCHR/HRW record, no anti-CCP inflation past the documents.

The spine is the Section 12 thesis — the US has assembled the same control surface on commercial-incentive logic and is one consolidation event away from operationally equivalent capability — which draws its US-side receipts from the body-layer references below.

Get updates on the Evil Robots series

Newsletter essays on AI escape, deception, and the humans who built them.