Category-level analysis of 100 statements across 5 sensitive content categories found that interface-based moderation gaps vary significantly by topic. Sexuality showed the strongest WebUI/API gap (WebUI 7.0× more likely to be moderated than API per GPT-4o judge for Gemini). Political ideology followed at 2.0×, then hate speech at 1.0×. Miscellaneous offensive topics showed the inverse pattern (API more moderated at 0.3×). Religious content showed WebUI moderation with no API moderation. The pattern suggests public-facing WebUI interfaces prioritize reputational risk management for high-scrutiny categories.

When the researchers attempted to use Gemini 2.5 Flash as a third independent LLM judge via its API for evaluating moderation decisions, Gemini automatically blocked all judging attempts citing safety reasons. This occurred even though the research task (judging whether a response is more or less moderated) does not itself produce harmful content. The incident illustrates that LLM safety systems can over-block legitimate research use cases, and that different LLM providers have different thresholds— Claude Haiku 4.5 and GPT-4o completed all judging tasks without safety refusals.

§3.3.3 detection

API and WebUI interfaces show statistically significant response length differences in opposite directions across models. Gemini API responses averaged 2,333 characters vs. 1,746 for WebUI (34% longer API; t=5.028, p<0.0001, Cohen's d=0.50). ChatGPT WebUI responses averaged 2,752 characters vs. 1,389 for API (98% longer WebUI; t=-9.800, p<0.0001, d=-0.98). The divergent direction across models suggests fundamentally different generation parameters rather than simple post-hoc filtering, indicating architectural or policy-level differences at the provider level.

§4.6 evaluation

An empirical study of 100 sensitive statements tested on Gemini (2.5 Flash) and ChatGPT (GPT-5) found that WebUI interfaces are systematically more restrictive than their API counterparts. According to GPT-4o judge: WebUI was moderated 18% of the time vs. 9% (Gemini API) and 13% (ChatGPT API). DeBERTa classifier found 82% of WebUI responses moderated vs. 58% of API responses. The Gemini WebUI:API ratio ranged from 2.0:1 (GPT-4o) to 7.0:1 (Claude), and ChatGPT from 1.4:1 (GPT-4o) to 15.6:1 (Claude). Neither Google nor OpenAI discloses these interface-specific policies.

§4.3, Table 2 evaluation

Across all 7 LLMs tested (GPT 4o, GPT 4o Mini, Gemini 1.5 Flash, Gemini 1.5 Pro, Llama 3.2, Claude 3.5 Haiku, Claude 3.5 Sonnet), statistically significant evidence of censorship bias was found in at least one evaluation metric per model: responses to Simplified Chinese prompts were more neutral, more similar to sanitized text, and less opinionated than semantically identical Traditional Chinese prompts (p < 0.05 across refusal-rate, sentiment, CensorshipDetector classification, and word-embedding analyses).

2025-ahmed-llm-censorship-bias Table 4, §6 keyword-filteringml-classifier

CensorshipDetector, an XLM-RoBERTa model fine-tuned on 587,819 Baidu Baike articles (censored) and Chinese Wikipedia (uncensored), achieved 91% accuracy on a held-out validation set of Chinese news articles, correctly classifying 93% of Chinese state media articles as censored and 87% of New York Times Chinese articles as uncensored, with average censorship scores of 0.93 and 0.13 respectively.

2025-ahmed-llm-censorship-bias §4.7, §4.7.2 keyword-filteringml-classifier

The paper proposes a black-box methodology for detecting censorship bias in LLMs by comparing responses to identical prompts in Simplified vs. Traditional Chinese — scripts for the same spoken language — controlling for translation quality while exploiting that Simplified Chinese training data is disproportionately sourced from mainland China's censored internet. Each prompt is repeated ten times and scored for similarity to censored text using an XLM-RoBERTa classifier fine-tuned on Baidu Baike (censored) vs. Chinese Wikipedia (uncensored) with scores from 0 to 1.

2024-ahmed-extended §2 / §2.4 keyword-filteringml-classifiermeasurement-platform

Across five popular translation services available in China (Alibaba, Baidu, Tencent, Youdao, and Microsoft Bing), researchers discovered 11,634 unique censorship rules in total. Every service — including the American-operated Bing Translate — implemented automatic censorship that silently omits content, with only Alibaba displaying any notification ('Query csi check not pass') to the user.

2024-ruo-lost §6 Results / Table 2 keyword-filteringml-classifier

Evidence from Youdao Translate suggests it deploys a machine-learning or NLP-based classifier alongside keyword rules: measured rules included repeated components (e.g., 螺+螺+螺+螺+螺+螺+蟢+D+哒+大) and nonsensical multi-token sequences that no human rule author would write, yet which consistently triggered censorship. Youdao returned 9,414 unique rules from the general test set — the most of any service — while also producing the most structurally anomalous rule patterns.

2024-ruo-lost §6 Results / §10 Future Work keyword-filteringml-classifier

China's local censorship operates through 'extra-institutional governance' (EIG) — practices that transgress the official identity and authorized means of the CAO system, including outsourced private surveillance, unpaid personnel secondment, and mass reporting via personal accounts — which upper-level offices tolerate but do not formally authorize, preserving plausible deniability when practices are ethically or legally questionable. The paper documents these as widespread implicit norms across China, not isolated to District T.

2024-zhang-toothless §Discussion and Conclusion keyword-filteringml-classifier