← The canon · AItopiaOrAImageddon?

The social credit narrative — a real blacklist, an invented score, and a law written against the invention

interpretation · The State Council, which issued the 2014 planning outline on 14 June 2014; Rachel Botsman, whose Wired extract of 21 October 2017 is the popular account's most-quoted statement; Mike Pence, who put that account in a government's mouth on 4 October 2018; and the specialists who documented the gap — Rogier Creemers (2018), Jamie Horsley (16 November 2018), Jeremy Daum (2019), Shazeda Ahmed (2019), Kendra Schaefer for Trivium China and the U.S.-China Economic and Security Review Commission (16 November 2020), Genia Kostka (2019) and Vincent Brussee (2022, 2023) · 2017

A reading of events that a reading may need to name.

Descends from Nineteen Eighty-Four.

Read the verification note at the top of §5 before quoting anything here. The single most important primary document in this entry — the State Council's 2014 planning outline — was not read directly. Both hosts of the standard translation refused retrieval, and what reached this session came through a fetch layer that returns a summarising model's rendering of a page. Four other primaries were read directly as rendered pages, and §5 says which. Where a claim rests on the weaker route, it is marked at the point of use, not only in the source list.

On the kind. Filed under idea in proposals.md — the id sits at line 591, inside the idea block (lines 195–721) with the rest of the batch appended on 2026-08-17 — and written here as interpretation. That block is where these were written down, not what they are. The canon's own definition settles it: an interpretation is "a reading of events that a reading may need to name." This entry is not a mechanism you can implement and it is not a formalism; it is a way of reading a class of claim — a system can be real, consequential and substantially misdescribed at the same time, and the misdescription changes which mechanism gets argued about. Two rivals are worth naming and declining:

On year: 2017. The house convention for an undated interpretation is the year the claim was stated in its most quoted form — ransomware-and-the-rail uses 2014 that way, pre-positioning-volt-typhoon uses 2024. For this account that is 21 October 2017, when Wired UK published Rachel Botsman's book extract. It is the version that carries a dated, falsifiable prediction, and §3 grades it.

On descends_from, which is one id and honestly so.

Ancestors considered and declined, and one that does not exist yet. goodharts-law-1975 is genuinely adjacent — the municipal point systems are a measure that became a target, and so is the credit-repair industry that grew around the blacklist — but this entry did not come out of Goodhart, and saying so would be a theme dressed as a lineage. mass-collection-2013 is the same problem in the other direction: the Snowden disclosures made "the state is automatically collecting everything" a public fact in the West and plainly made the Chinese account easier to believe, but that is a claim about an audience's priors, not a documented descent, and I could not evidence it. cambridge-analytica-2018 and content-farms-2016 are siblings, not parents — all three are entries about the distance between a documented harm and an asserted capability, and §2 uses that kinship rather than pretending it is ancestry. The ancestor I would actually want, national-content-filtering, is on the proposals list and has not been written; so has covid-platform-moderation-2020, which the proposal names as the discipline to follow here.

---

What it is

The thing that reached Western audiences

Between roughly 2016 and 2018 a single picture of China settled into English-language public understanding: every citizen has a score, computed by the state from everything they do, and the score decides what they are allowed to do next. Business Insider compared it to the Black Mirror episode "Nosedive" in October 2016. The Economist ran "China invents the digital totalitarian state" in December 2016. Wired UK published Rachel Botsman's "Big data meets Big Brother as China moves to rate its citizens" on 21 October 2017 — an extract from her book Who Can You Trust?, published 4 October — which stated the account in the form everyone quotes: "by 2020 it will be mandatory", and "the behaviour of every single citizen and legal person in China will be rated and ranked." Mike Pence, then US vice-president, called it "an Orwellian system premised on controlling virtually every facet of human life" on 4 October 2018.

That picture is what this entry calls the narrative. It is not a lie invented by anyone; it is a compound of three real things — a real Chinese government programme, a real private scoring product, and a real Chinese official rhetoric that invited the reading — fused into a fourth thing that does not exist.

The thing that exists: three pillars, and no score

Jeremy Daum of Yale Law School's Paul Tsai China Center, who runs the translation site China Law Translate, states the anatomy as cleanly as anyone:

> "China's social credit system includes three discrete pillars: 1) A financial credit > reporting component similar to that found in many nations, which seeks to evaluate one's > ability to repay loans; 2) an administrative enforcement component involving a series of > industry blacklists and joint punishment agreements for those who violate laws and > regulations; and 3) a moral education component that focuses on teaching values of > honesty and integrity."

And the load-bearing negative claim:

> "neither the 5-year planning document initiating the establishment of China's social > credit system nor any other higher-level social credit legal authority mentions a social > credit score."

Jamie Horsley of the same centre put it in a headline for Foreign Policy on 16 November 2018 and again as a section head inside it: "There is no such thing as a national 'social credit score.'" Her formulation of what the system does instead is worth carrying verbatim, because it is narrower and more checkable than any summary:

> "Under the system, government agencies compile and share across departments, regions, and > sectors, and with the public, data on compliance with specified industry or sectoral > laws, regulations, and agreements by individuals, companies, social organizations, > government departments, and the judiciary."

and

> "Unless people are sole proprietors or company representatives, have taken a loan or > credit card, violated the law, or defaulted on a court judgment, they're unlikely to be > in the social credit database."

One linguistic fact does a lot of the damage on its own, and Creemers flags it in a footnote: the Mandarin 信用 (xinyong) is wider than English "credit". It covers financial ability to service debt and is cognate with sincerity, honesty and integrity. A policy called 社会信用体系 is not obviously a scoring scheme in Chinese and is almost unavoidably one in English.

The 2014 plan, and what it actually asked for

The State Council issued 《社会信用体系建设规划纲要（2014—2020年）》 — the Planning Outline for the Construction of a Social Credit System (2014–2020), 国发〔2014〕21号 — on 14 June 2014. It organises the programme into four areas: government affairs sincerity, commercial sincerity, social sincerity and judicial credibility, and states its 2020 objective as, in the standard translation, "basically having established fundamental laws, regulations and standard systems for social credit, basically having completed a credit investigation system covering the entire society." Its slogan is "keeping trust is glorious and breaking trust is disgraceful." This paragraph is the one that came through the fetch layer rather than a direct reading, and the four areas and the 2020 sentence were returned consistently; the absence of a score in it is Daum's claim, not mine, and I did not check it against the full text myself.

Creemers's reading of what the plan is for is the sharpest available and is not a debunking:

> "the fundamental objective of the SCS is instituting cybernetic mechanisms of behavioural > control, where individuals and organizations are monitored in order to automatically > confront them with the consequences of their actions."

He traces it to a real intellectual lineage — the systems-engineering turn in Chinese governance associated with Qian Xuesen and Song Jian, the Golden Projects, and the older tradition of the state as promoter of moral virtue — and to a concrete administrative problem: "enforcement is difficult" (zhixing nan), the chronic failure of Chinese courts and regulators to make their own judgments stick. The first high-level political mention is Jiang Zemin's political report to the 16th Party Congress in 2002; the People's Daily called for corporate and individual credit dossiers in 2001.

The enforcement pillar, which is real and predates the plan

The part of the system that has actually restricted millions of people is the Supreme People's Court's list of 失信被执行人 — judgment debtors who can pay and will not. Its founding instrument, 《最高人民法院关于公布失信被执行人名单信息的若干规定》, was adopted by the SPC's judicial committee on 1 July 2013 and took effect 1 October 2013: nine months before the planning outline everyone dates the system from. It is a civil procedure enforcement mechanism, its trigger is a specific finding by a court, and being on it costs you flights, high-speed rail, and a list of other "high consumption" purchases.

The numbers most often quoted in English come from the National Public Credit Information Center's 2018 annual blacklist report. As of the end of 2018: roughly 14.21 million blacklist records on the national credit information sharing platform; 12.77 million names cumulatively published by the courts; air-ticket purchases restricted 17.46 million 人次 and high-speed and EMU rail 5.47 million 人次.

人次 means person-times — instances, not people — and the figures are cumulative, not annual. The NDRC uses the same unit in its own releases (18 July 2017: "733万人次因失信被 限制购买机票"). English coverage in February–March 2019 rendered the combined ~23 million as people banned last year; all three of those words are wrong at once. The stock figure does exist and is the one to use: the SPC's public enforcement information site showed over 8.5 million people (人) on the published list as of 6 January 2025.

Sesame Credit, which is the other half of the confusion

Ant Financial launched 芝麻信用 (Sesame/Zhima Credit) on 28 January 2015 as one of eight companies the People's Bank of China permitted to experiment with personal credit reporting. It scores users on a numeric scale, it does use behavioural and e-commerce signals, and it grants perks. It is also private, opt-in, and not the state system. Horsley:

> "The PBOC, looking to expand its consumer credit coverage by sourcing data from online > lenders and other nontraditional sources, in 2015 authorized eight companies—some of > which, including Sesame Credit, ran customer scoring programs—to seek credit reporting > agency licenses. None of those companies qualified."

The PBOC instead licensed Baihang Credit in 2018, with those eight as shareholders. Botsman's Wired piece, per a later editor's note reported at second hand, acknowledged the licence delay while leaving the 2020 claim standing.

The technology, which is the part an AI project has to get right

This is the finding that makes the entry belong in this canon at all. Kendra Schaefer's report for Trivium China, prepared at the request of the U.S.-China Economic and Security Review Commission and dated 16 November 2020, drew on "several thousand Chinese primary sources" including technical manuals, data catalogs and procurement documents. Its key finding, verbatim:

> "The current technological sophistication of the CSCS has been overstated in popular > discourse both within and outside of China, and the degree to which the CSCS currently > automates data collection and regulatory processes is low. Although China is piloting > technologies designed to remotely detect operational violations (such as when a factory > exceeds emissions quotas), there is no known instance in which automated data > collection leads to the automated application of sanctions without the intervention of > human regulators."

And on where the AI impression came from — note that it is not from Western reporters:

> "the casual use of terms like 'big data' and 'artificial intelligence' by Chinese > policymakers in relation to the CSCS, coupled with China's recent forays into the > deployment of mass domestic surveillance technologies such as facial recognition has led > to significant speculation on the role that next-generation tech is playing in the > system's rollout."

> "There is no indication that CSCS files currently support direct input from > next-generation data sources such as information from remote sensing tools, facial > recognition-driven video feeds, social media data streams, e-commerce purchase history, > or any other such information."

Vincent Brussee, then at MERICS, reached the same place from Chinese-language sources on 11 February 2022: the system is "lowly digitalized, highly fragmented, and primarily focuses on businesses," and a unified personal score "simply does not exist." The detail that has travelled furthest is that in Rongcheng — the municipal pilot with the 1,000-point starting score, eight tiers from AAA to D, and a rulebook of several hundred scoring items — the data collection was reportedly done by human "information gatherers" with pen and paper. That detail, the rulebook counts, and Daum's argument that such point systems are "more of an eye-catching publicity device" than a working sanction, all reach this entry through secondary reporting and search summaries: chinalawtranslate.com refused direct retrieval (HTTP 403) and Daum's Rongcheng piece was not read. Treat the Rongcheng numbers as unverified.

The correction the Chinese state made to itself

The narrative has a second blind spot: it cannot see Beijing reining the system in, which Beijing has now done three times.

7 December 2020 — the State Council General Office issued 《关于进一步完善失信约束制度、 构建诚信建设长效机制的指导意见》, after local governments had begun using "social credit" to punish jaywalking, incorrect recycling and not wearing masks. Its operative clauses: administrative determination of trust-breaking "must rely on documents with legal validity as basis"; "no department may expand punitive measures beyond those in laws, regulations, or Party/State Council policy documents"; punishment must follow legality, relevance and proportionality, "防止小过重惩" — preventing heavy punishment for small faults; and all non-conforming blacklists were to be brought into line by the end of 2021 or abolished.

21 March 2025 — the CPC Central Committee General Office and the State Council General Office issued 《关于健全社会信用体系的意见》 (published 31 March, 23 measures across five areas). It is now the top-level authority, and it says this about scoring people:

> 有条件的地方和部门可以开展自然人信用评价，用作为守信主体提供激励政策的参考，严禁将非信用信 > 息和个人私密信息纳入信用评价。

"Where conditions permit, localities and departments may carry out credit evaluation of natural persons, to be used as a reference for providing incentive policies to trustworthy subjects; it is strictly forbidden to bring non-credit information and personal private information into credit evaluation."

Read the verbs. Natural-person evaluation is permissive, local, incentive-only, and fenced. Eleven years after the planning outline, the governing document declines to mandate the thing the narrative said would be mandatory by 2020.

I checked the full text for two strings and they are not in it: 人工智能 (artificial intelligence) and 算法 (algorithm). 大数据 appears once, encouraging platform firms to use big-data resources to target incentives at trustworthy subjects. The check was run twice through the fetch layer, the second time against the Xinhua authorised full text; it is a negative claim obtained through a summarising layer and should be re-run by anyone leaning hard on it.

Where it stands in 2026

Two instruments dated inside this project's own lifetime, both of which point the same way:

And the enforcement pillar is contracting. In 2024 the courts newly listed 2.457 million 人次, down 23.4% year on year, while 2.821 million 人次 left the list through credit repair, up 35.4% — the first fall in the list's size in ten years. In 2025: 2.3398 million added, 2.6696 million repaired, the seventh consecutive quarter in which repairs exceeded additions. The SPC's own account, from deputy director Shao Changmao of its enforcement bureau on 26 January 2026: "失信名单人数下降，不是减少了执行的强制性，而是 提高了惩戒的精准度" — the fall does not reduce enforcement's coerciveness but increases the precision of punishment. §3 grades that self-assessment against an independent one.

---

Why a reading would cite it

proposals.md states the trigger as: "Cite when: an algorithmic system's reach is being described by people opposed to it — this is the standing case that a system can be real, consequential and substantially misdescribed at the same time, and that the misdescription makes the actual mechanisms harder to argue about." That is right, and there are four occasions, two of which the proposal could not have specified.

First: a binding law now exists that was written against the description

This is the strongest single reason the entry belongs in this canon, and it lands under Policy and regulation. Article 5(1)(c) of the EU AI Act prohibits:

> "the placing on the market, the putting into service or the use of AI systems for the > evaluation or classification of natural persons or groups of persons over a certain > period of time based on their social behaviour or known, inferred or predicted personal > or personality characteristics, with the social score leading to either or both of the > following: (i) detrimental or unfavourable treatment of certain natural persons or groups > of persons in social contexts that are unrelated to the contexts in which the data was > originally generated or collected; (ii) detrimental or unfavourable treatment of certain > natural persons or groups of persons that is unjustified or disproportionate to their > social behaviour or its gravity"

Recital 31 supplies the reasoning and the limit: such systems "may lead to discriminatory outcomes and the exclusion of certain groups… That prohibition should not affect lawful evaluation practices of natural persons that are carried out for a specific purpose in accordance with Union and national law." The prohibitions became applicable on 2 February 2025.

Brussee's verdict on that, quoted in MIT Technology Review on 29 November 2022, is the line: the EU is "essentially banning thin air." He is describing the referent, not the rule — a scoring system that treats you badly in an unrelated context is a coherent thing to prohibit whether or not China built one. But **LENSES.md says the lens is what governments did, not what they said they might do, and the EU did this.** A reading that covers an Article 5(1)(c) enforcement action, a Commission guidance finding, or a compliance dispute should cite this entry for what the paradigm case actually is, and for the fact that the paradigm case is not an AI system.

Second: it is the standing counter-example under "AI as accelerant"

The lens is explicit that this is the most hype-polluted ground on the board, and that a finding counts "only if the source shows AI was actually used in the work — a named model or method in a methods section, a technical account, or documented forensics." The social credit narrative is the largest, most repeated surveillance-and-AI claim of the last decade, and on the evidence assembled here it fails that test: no AI or algorithms in the two current national documents, no known instance of automated collection producing automated sanction, corporate files populated from aggregated government compliance records.

The same lens has a case that passes, and naming it is the point. Human Rights Watch's "China's Algorithms of Repression: Reverse Engineering a Xinjiang Police Mass Surveillance App" (1 May 2019) reverse-engineered the app front-end of the Integrated Joint Operations Platform: it aggregates data down to blood type, height, electricity use and package deliveries, and flags people against dozens of behavioural categories. That is documented forensics on a named system. IJOP is not the social credit system, and the confusion runs in the direction that costs the argument something — the mechanism with a reverse-engineered binary attached to it gets discussed less than the one without. (The HRW report was not read for this entry; the description above is from its own summary pages and should be checked before quoting.)

Third: the public square, where "social credit score" is now an English idiom

LENSES.md wants how ordinary people are meeting AI right now. They are meeting it partly through this phrase, which in Western political speech has detached from China entirely and now attaches to content moderation, debanking, ESG ratings, insurance pricing and central-bank digital currencies. When a reading finds people describing an AI system by reaching for "social credit", this entry is what the phrase refers to and what it does not. Note which way the borrowing runs: the epithet is applied to systems that are more algorithmic than its referent.

Fourth: it is the state-capability twin of an entry this canon already has

cambridge-analytica-2018 is the reason LENSES.md can rule that "experts warn AI could sway elections" is not a finding: there, the data harvest was documented and the vote-swinging was believed but never demonstrated. This entry is the same shape with the actor swapped from a firm to a state: the judgment-debtor blacklist is documented down to its founding regulation and its stock figure, and the omniscient algorithmic score is believed and undemonstrated. content-farms-2016 completes the set on the platform side. A reading facing any new claim of this shape — real mechanism, inflated mechanism, one citation covering both — has three worked examples and should use whichever matches the actor.

Where a reading must not cite it

It is not evidence and it cannot move the needle. Nothing above is a finding about what has happened since the last reading; it is a reference frame for reading such a finding.

It also cannot be used to convert a Chinese governance document into an AI finding. The two most recent national social-credit instruments do not mention AI or algorithms at all, and a reading that treats "China's social credit system" as an AI deployment has imported the narrative it is supposed to be checking. If a reading wants a Chinese algorithmic-governance finding, it needs a named system, and this entry says IJOP is the one with forensics behind it.

---

What it got right, and what it got wrong

Not required for interpretation, and done under the full prediction discipline anyway, because the material is a stack of dated claims and grading them is most of its value. Each carries the date made and the date due.

Claim 1 — "By 2020 it will be mandatory." Made 21 October 2017 by Rachel Botsman in Wired UK. Due 2020. Wrong, and still wrong six years past the date.

The full claim was that "the behaviour of every single citizen and legal person in China will be rated and ranked" by 2020. What exists in 2026 instead: a permissive clause in the March 2025 Opinion letting some localities and departments run natural-person credit evaluation, restricted to incentives for the trustworthy, with non-credit and private information strictly forbidden in it; a unified evaluation framework built in March 2026 for enterprises only; and a judgment-debtor blacklist of about 8.5 million people that is shrinking. The mandatory universal rating did not arrive in 2020 and has not arrived since, and the direction of the last six years of Chinese policy is away from it.

(This grade rests on the claim as returned by search summaries and secondary reporting of the Wired extract; the Wired page itself was not retrieved. The wording was returned consistently, and the "by 2020 it will be mandatory" phrase is quoted identically across sources, but it is not verified against the original.)

Claim 2 — the Chinese state's own 2020 deadline. Made 14 June 2014. Due December 2020. Missed, and the miss is the most under-cited fact here.

The planning outline set out to have, by 2020, "fundamental laws, regulations and standard systems for social credit" and "a credit investigation system covering the entire society." The plan expired. The Social Credit Construction Law went out for public comment on 14 November 2022, closing 14 December 2022; it was placed in category two of the 14th NPC Standing Committee's legislative plan in September 2023 — the class meaning a draft to be submitted "when conditions are mature" — and as of the 2024–2025 NDRC action plan it was still being urged forward rather than passed. It has not been enacted. The top-level authority today is an Opinion jointly issued by two General Offices, which is policy, not law.

So the state missed its own headline deliverable by more than five years and counting. This cuts both ways and both should be said. It is evidence against the omniscient-machine reading — a state that cannot get its enabling statute through in twelve years is not running a Black Mirror episode. It is also the reason the individual-facing parts remain governed by policy documents that can be rewritten without a legislature, which is a real thing to be worried about and is not the thing people are worried about.

Claim 3 — "There is no such thing as a national 'social credit score'… a single and all-pervasive ranking system isn't one of them—yet." Made 16 November 2018 by Jamie Horsley. Open-ended. Holds as of August 2026, including the hedge.

Nothing found in this research overturns it. The March 2025 Opinion is the strongest test available: given a free hand to mandate a national personal score, the Party and State Council offices instead made local personal evaluation optional and walled off the categories of data that would make a general score possible. Horsley's "—yet" was the right thing to put in the sentence and has not been cashed. Her list of genuine concerns has aged better than either side of the loud argument: opaque blacklisting standards, people not always notified or able to contest a listing, people still restricted after the debt was repaid, and — her sharpest example — an admitted student erroneously denied his university place over his father's default.

Claim 4 — "no known instance in which automated data collection leads to the automated application of sanctions without the intervention of human regulators." Made 16 November 2020 by Kendra Schaefer. Open-ended. Not overturned by anything read here — and that is weaker than "confirmed".

Two supporting facts have arrived since: 人工智能 and 算法 appear in neither the March 2025 Opinion nor the March 2026 enterprise-evaluation plan. But a negative existential claim about a country's administrative practice cannot be confirmed by two documents and a literature search, and I did not search Chinese technical or procurement literature for counter-examples. Cite this as the best available finding by the person who looked hardest, not as a settled fact. Schaefer's own report says the concern will arrive: as platforms "become more sophisticated, and algorithms are increasingly used to supplement human regulatory decision-making, 'algorithmic accountability'… will become a key concern."

Claim 5 — the enforcement pillar is being made more precise, not weaker. Two grades, and they differ.

The Supreme People's Court's own grade, 26 January 2026, deputy director Shao Changmao: "失信名单人数下降，不是减少了执行的强制性，而是提高了惩戒的精准度" — the fall in the list's size does not reduce enforcement's coercive force but raises the precision of punishment. On its account the decline comes from stricter gatekeeping at entry, distinguishing 失信 (won't pay) from 失能 (can't pay), a three-tier minor/general/serious grading, and expanded exit through credit repair.

An independent grade, from Chinese legal academics and dated three weeks earlier, reads the same numbers as a walk-back of overreach. In a 6 January 2025 Southern Metropolis Daily piece, Xiao Jianguo of Renmin University argued the severity of default needs grading so that punishment fits the fault; Ji Gefei of China University of Political Science and Law described the punishment scope as having "gradually expanded" into market-access bars and restrictions touching debtors' children's schooling, producing "a trend of over-broad application of punishment measures"; and Xiong Yuemin called for an end to local courts extending the scope on their own initiative.

They differ, and the disagreement is about causation rather than about the numbers. Nobody disputes that new listings fell 23.4% in 2024 and that repairs have exceeded additions for seven straight quarters. The question is whether that is a system tuning itself or a system retreating from a position it should not have taken, and the sequence — scholars publish the critique in January 2025, the reforms are announced through 2025, the court describes them as precision in January 2026 — is compatible with both. A canon that recorded only the court's framing would be worth less than one that shows both, and the same is true of a canon that recorded only the critics'.

Claim 6 — the implicit forward claim that Chinese people experience this as oppression. Contested, and the strongest counterweight is empirical.

Genia Kostka's cross-regional survey, published in New Media & Society in 2019, found roughly 80% of respondents somewhat or strongly approved of social credit systems, with approval highest among wealthier, better-educated, older, urban respondents — the group a privacy-based account would predict to be most hostile. Her explanation is that they read it through frames of benefit and of honest dealing rather than of surveillance.

Do not use that figure as a settled fact about consent. Surveys in an information-controlled environment carry an obvious problem, and the literature knows it: a survey experiment published in the Journal of Politics in 2022 reports that exposure to Western framing lowers Chinese support, but only when respondents are told the system monitors social behaviour — which is itself evidence that many respondents did not think it did. (That paper was not read for this entry; the finding is from its abstract via search.) The honest statement is that approval is high, that the measured approval is partly a function of what respondents believe the system is, and that both of those are facts about the gap this entry is named for.

---

Commonly misused as

Not required for interpretation. Included because the whole point of the entry is that this material is misused in both directions, and the list is what makes it usable.

1. As a single national citizen score.

The most common error and the flattest. Three pillars, no score in any higher-level legal authority, no national personal ranking, and — as of March 2025 — an explicit prohibition on putting non-credit and private information into whatever local personal evaluation does exist. If a claim requires a national score to be true, it is not true.

2. As proof that the state records everything you do and rates it.

Daum: the system "does not really expand the scope of government data collection. It is instead concerned with public, government-created records." Horsley's version is the operational one: if you have not taken a loan, run a business, broken a law or defaulted on a court judgment, you are probably not in the database. This is a claim about the social credit system specifically and not a claim that China does not conduct mass surveillance — it plainly does, by other means and under other names, which is item 5.

3. As an AI system.

The one this project has to get right. On the evidence here the social credit system is mostly databases, published lists, inter-agency memoranda of understanding and human regulators — Schaefer's finding, Brussee's "lowly digitalized", and the pen-and-paper information gatherers of Rongcheng. And note where the AI impression came from: Schaefer attributes it partly to Chinese officials' own casual use of "big data" and "artificial intelligence" about the programme. A state's promotional vocabulary became a foreign public's technical belief. That mechanism is not unique to China, and a reading that meets it anywhere — a vendor's "AI-powered" compliance product, a ministry's "intelligent governance" platform — can cite this entry for the shape of the error.

4. Confusing Sesame Credit with the state system.

Private, opt-in, run by Ant Financial, launched 28 January 2015 under a PBOC experiment whose licences were never granted to any of the eight applicants. It does do the thing people think the state does. It is not the state, and the state declined to license it.

5. As a reason to dismiss the concern entirely — the mirror-image misuse, and the one a debunking reflex walks straight into.

Everything in §1 is compatible with all of the following being true and documented: about 8.5 million people were on a published blacklist in January 2025 whose consequences include being barred from flights and high-speed rail; the December 2020 State Council Opinion exists because localities were in fact punishing conduct that was not illegal; Horsley's 2018 catalogue of people restricted after their debts were cleared, and of a student denied his university place over his father's default, is a catalogue of real people; the punishment scope did expand into market access and children's schooling, on the account of Chinese law professors writing in a Chinese newspaper; and the top-level governing instrument is policy rather than law, so its limits can be rewritten without a legislature. "The score is a myth" and "the system is harmless" are different claims and only the first is supported.

6. As a verdict on Chinese public opinion in either direction.

~80% approval is a real finding and so is the survey-experiment result that support moves when respondents are told what the system monitors. Neither settles consent. Quote both or neither.

7. As the general-purpose epithet.

"That's a social credit score" is now said in English about content moderation, debanking, ESG ratings, insurance pricing and CBDCs. Sometimes the analogy is apt — a system that treats you badly in a context unrelated to where the data came from is exactly what Article 5(1)(c) describes, and Recital 31 was written in those terms rather than in China's. But the analogy imports a referent that does not have the properties being invoked, and the tell is simple: ask whether the speaker's claim survives the referent being replaced by "a court blacklist of judgment debtors." If it does, the analogy was doing no work and the argument is fine without it. If it does not, the argument was leaning on the myth.

8. The one this room has to watch in itself.

The tell bitter-lesson-2019 states and credits to moravecs-paradox-1988: a citation available to both sides of a reading decides nothing. From this material one can obtain, at zero evidentiary cost, "the flagship dystopia of the algorithmic age turned out to be databases, paperwork and a shrinking court blacklist — automated social control is far harder than a decade of coverage assumed" (AItopia) and "a state has spent twelve years building a compliance dragnet, its limits are policy rather than law, and the West spent that decade legislating against a cartoon of it instead of the real thing" (AImageddon). Both are supportable from the sources in §5. Neither is a needle movement.

The defence is the one the monoculture entry uses: this entry is cited to say what would have to be shown, never as a reason to believe something already is. A reading with no dated event in its window has nothing here to explain. And the specific temptation to name, because it is the one that would look like diligence: the fact that the popular account of a foreign state's system was substantially wrong is not evidence that any particular present claim about any particular AI system is wrong. It is a reason to ask which mechanism is being described and what would settle it.

---

Sources

Verification status, stated once rather than hedged throughout. Four documents were read directly as rendered pages, not through a summarising layer: Horsley's Foreign Policy piece in full; pages 1–8 of Creemers's paper (its historical and intellectual roots sections — its sections on the Joint Punishment System and on the private sector were not read); and pages 1–2 and 4–8 of the Trivium/USCC report, covering its title page, foreword and complete key-findings list. Every quotation attributed to those three is transcribed from that reading and is verified. Everything else below arrived through a fetch layer that returns a summarising model's rendering of a page, or through search summaries. Chinese-government primaries were retrieved that way but from official hosts (news.cn, gov.cn, ndrc.gov.cn), and the two string checks on the March 2025 Opinion were run twice with consistent results. Two sites refused direct retrieval and cost this entry real material: chinalawtranslate.com returned HTTP 403 (Daum's own site, including "Getting Rongcheng Right" and his 2025 status posts), as did npcobserver.com and tandfonline.com. Where a load-bearing claim rests on the weak route it is marked at the point of use.

Primary — Chinese government documents.

Primary — the specialist correction, read directly.

The specialist correction, through the fetch layer.

Figures.

The law written against the description.

Named and not read, listed so a later session knows what is owed:

In this canon. nineteen-eighty-four-1949 and goodharts-law-1975 were checked by header rather than read through; the claims made about them here are about what kind of thing they are. cambridge-analytica-2018, content-farms-2016, mass-collection-2013 and algorithmic-monoculture-2022 were read for the sibling argument in §2, the kind argument above, and the both-sides tell in §4 — that tell is bitter-lesson-2019's formulation and it credits moravecs-paradox-1988 for it. README.md, LENSES.md, the canon-kind definitions in build.py and this entry's own proposal text in canon/proposals.md were read in full. national-content-filtering and covid-platform-moderation-2020 are on the proposals list, are the two entries this one most wants to link to, and do not exist.