China-originating LLMs showed higher refusal rates, shorter responses and more inaccurate answers on 145 China-politics questions than non-China models; the study is observational and does not identify regulation as the sole cause.

SocioLens project,

Claim LD10-HQ-C23 (v1.0)

This is an older version. Read the current version, v1.1, or what changed in it.

China-originating LLMs showed higher refusal rates, shorter responses and more inaccurate answers on 145 China-politics questions than non-China models; the study is observational and does not identify regulation as the sole cause.

Provenance and status

Claim type: Estimate
Publication status: Eligible for publication, with qualifiers: non causal
Verification state: Checked against the primary source. Who checked this claim is not recorded for it; claim checks so far were AI-led, and no person has checked claims against their sources.
Does not license: Causal regulatory effect or downstream belief effect
Limitations: Observational; model vintages; provider and technical differences
Source (as cited): Pan & Xu, PNAS Nexus 2026
Source type: Peer-reviewed journal article (recorded as “Peer-reviewed article”)
Institutional position: Independent academic
Maturity of the evidence: Emerging evidence
Recorded date: 25 September 2026
Currentness / correction: This is not the current version. No correction or concern is recorded on its sources.

Sources this version cites

Pan & Xu, “Political censorship in large language models originating from China,” PNAS Nexus 5(2):pgag013 (2026) (Peer-reviewed journal article)
Read the original

Version

This page shows version v1.0. Its record never changes, even when newer evidence arrives; the site may add a note to the page or withhold a figure not cleared for public use.

NEWER VERSION EXISTS: read the current version, v1.1 (LD10-HQ-C23--v1.1). This version is explicitly superseded by LD10-HQ-C23--v1.1.

Versions: v1.0, v1.1 · current

Raw record: every recorded field, with its coded values

All 25 canonical Claim fields, verbatim, with the coded values derived from them. The plain-language labels above are read from these.

Claim type (coded): ESTIMATE (as printed: “ESTIMATE”)
Publication status (coded): ELIGIBLE (NON_CAUSAL) (as printed: “ELIGIBLE WITH NON-CAUSAL QUALIFIER”)
Verification state (coded): CONTENT_VERIFIED_PRIMARY
Source type (coded): PEER_REVIEWED_JOURNAL_ARTICLE
Maturity (coded): EMERGING

#	Canonical field	Canonical value (verbatim)	Normalized (Claim Value Vocabulary v1.0.2)
1	CLAIM ID	LD10-HQ-C23	—
2	PUBLIC-FACING CLAIM	China-originating LLMs showed higher refusal rates, shorter responses and more inaccurate answers on 145 China-politics questions than non-China models; the study is observational and does not identify regulation as the sole cause.	—
3	CLAIM TYPE	ESTIMATE	base_types: ESTIMATE
4	RESEARCH STRAND	F — Platform/jurisdiction governance	—
5	MATURITY	EMERGING	states: EMERGING
6	SOURCE	Pan & Xu, PNAS Nexus 2026	source_ids: SRC-LD07-PAN-XU-CENSORSHIP-2026
7	SOURCE TYPE	Peer-reviewed article	classes: PEER_REVIEWED_JOURNAL_ARTICLE
8	INSTITUTIONAL POSITION	Independent academic	—
9	SCOPE	Chinese- and English-language prompts; China-origin vs non-China models	—
10	DESIGN	Cross-sectional audit	—
11	SAMPLE / DATASET	145 political questions plus less-sensitive set	—
12	MODEL / TOOL / VERSION	BaiChuan, ChatGLM, Ernie, DeepSeek vs Llama/GPT families	—
13	COMPARATOR	China-origin vs non-China-origin models	—
14	MAIN RESULT	Large refusal/content differences; smaller gaps on less-sensitive prompts; all models refused more in Chinese	—
15	INFERENTIAL TARGET	Association between provider origin/jurisdiction and model behaviour	—
16	DOES NOT LICENSE	Causal regulatory effect or downstream belief effect	—
17	VERIFICATION STATE	CONTENT VERIFIED	state: CONTENT_VERIFIED_PRIMARY · verified_by: UNKNOWN
18	LIMITATIONS	Observational; model vintages; provider and technical differences	—
19	CONTRADICTORY / SUPPORTING SOURCES	Chinese regulatory framework is a plausible institutional condition	—
20	GEOGRAPHY / LANGUAGE	PRC-origin vs non-PRC / Chinese and English	—
21	DISTRIBUTIONAL LEDGER	Chinese-language users received more refusals across models	—
22	R01 / FIELD-MAP LINKS	D7, D6, D9, L-B, R-1	—
23	WHAT WOULD CHANGE THIS CLAIM?	Newer-model longitudinal audit or causal design	—
24	PUBLICATION STATUS	ELIGIBLE WITH NON-CAUSAL QUALIFIER	state: ELIGIBLE · qualifiers: NON_CAUSAL
25	RECORDED DATE	2026-09-25	—

Field 22 references resolved against Field Map v0.2: D7 (DOMAIN); D6 (DOMAIN); D9 (DOMAIN); L-B (LENS); R-1 (LENS)

The sources as recorded, with their record IDs and coded source types:

Pinned Sources (field 6, version-specific)

“Pan & Xu, PNAS Nexus 2026” → Pan & Xu (2026), “Political censorship in large language…” (SRC-LD07-PAN-XU-CENSORSHIP-2026--v1.0) — Pan & Xu, “Political censorship in large language models originating from China,” PNAS Nexus 5(2):pgag013 (2026); class PEER_REVIEWED_JOURNAL_ARTICLE (explicit public provenance pin)

Cited by Living Debate versions

Persuasion, Intermediation & the Information Order (v0.1) LD-10

Cite this page

SocioLens project. (2026, September 25). China-originating LLMs showed higher refusal rates, shorter responses and more inaccurate answers on 145 China-politics questions than non-China models; the study is observational and does not identify regulation as the sole cause [Claim LD10-HQ-C23, version v1.0]. SocioLens. https://sociolens-web.pages.dev/claims/ld10-hq-c23/v1.0/

@misc{sociolens_ld10_hq_c23_v1_0,
  author = {{SocioLens project}},
  title = {China-originating LLMs showed higher refusal rates, shorter responses and more inaccurate answers on 145 China-politics questions than non-China models; the study is observational and does not identify regulation as the sole cause.},
  year = {2026},
  date = {2026-09-25},
  howpublished = {\url{https://sociolens-web.pages.dev/claims/ld10-hq-c23/v1.0/}},
  note = {Claim LD10-HQ-C23, version v1.0}
}

The link points to this version, whose record never changes. For a finding, cite the original study as well: SocioLens summarizes research and does not replace it.

Research-of-record provenance

Materialized by sociolens-materializer 0.6.0 at 2026-09-28T20:00:00.000Z; normalization contract SocioLens — Claim Value Vocabulary v1.0.2 — Pre-Spike Materialization Contract (v1.0.2). Content seal SHA-256 ccce023c1cf6e90f337f8c78fd99b0634f60152d96715db00d9a660a0975540b.

Frozen Drive-derived inputs

R07 — LD-10 Canonical Claim Ledger v1.0 — LOCKED (v1.0) — Drive ID 1f9tgGnqv80_UyhW0kCgcnTgWNPlx-DYM; revision: NOT_EXPOSED_BY_DRIVE_TEXT_EXPORT; Drive modified ; block “ENTRY 23 — LD10-HQ-C23 ” SHA-256 1c15529588e6e90debe5f441b286790736fa6b3be5bf55539cf206aa29e844e2; SocioLens — Programme Status & Authority Register v0.6 status “CURRENT.”; frozen fixture fixtures/frozen/r07-ld10-claim-ledger-v1.0-canonical.

Builder decisions (not Drive data)

VERSION = v1.0 — basis: Claim version taken from the canonical ledger artifact version (R07 — LD-10 Canonical Claim Ledger v1.0); Implementation Architecture Selection §5 logical_id/version convention.
CURRENT_POINTER = false — basis: not current: superseded by LD10-HQ-C23--v1.1 (manifest supersedes relation)
FIELD_6_SOURCE_REFERENCE = SRC-LD07-PAN-XU-CENSORSHIP-2026--v1.0 — basis: field 6 of ENTRY 23 — LD10-HQ-C23 cites "Pan & Xu, PNAS Nexus 2026"; Builder identification: the cited work is the already-materialized Source record SRC-LD07-PAN-XU-CENSORSHIP-2026--v1.0; exact identity established by authority, not inferred (Issue #53 Source inventory, factory-hq decision HQ-ESC-sociolens-ld10-t5-20261011T050405Z-01 ruling 3 inventory U1 / item 7(d)): R07 Batch 03 §4 ("Primary PNAS Nexus article checked: Pan & Xu, 2026") and R03 Batch 04 §1 (the identity authority of the existing record) name the same work; SocioLens record v1.0 is its sole record and no second identity is created.
SOURCE_PROVENANCE_PIN = SRC-LD07-PAN-XU-CENSORSHIP-2026--v1.0 — basis: Issue #5 proof rendering (Architecture Proof Protocol v0.1 AP-08 / Amendment v1.2 §6): explicit version-pinned public provenance of this ClaimVersion — the pinned field-6 Source record "Pan & Xu, PNAS Nexus 2026"; reuse of the already-materialized record by authority (Issue #53 Source inventory, factory-hq decision HQ-ESC-sociolens-ld10-t5-20261011T050405Z-01 ruling 3 inventory U1 / item 7(d)).