related work

← All pages

This page summarizes the peer-reviewed and preprint literature in sources/peer_rev_papers/ that informs this wiki’s methodology, citations, and framing. Every paper below was verified by reading its actual title page, abstract, and author list — not assumed from filename or topic — before being cited anywhere in this wiki.

Note on scope: sources/peer_rev_papers/ contains 6 PDFs. Each is summarized below. No other papers exist in this folder; this wiki does not cite any paper it cannot trace back to an actual file in this source directory.

Davidson, Muir, Burnat & Joinson (2026)

File: 2601.08415v2.pdf · arXiv:2601.08415v2 [cs.CY]

A comparative analysis of the Terms of Service of five major LLM providers — Anthropic, DeepSeek, Google, OpenAI, and xAI — collected November 2025, identifying “regulatory gray areas” where Terms create uncertainty for legitimate research use (security research, computational social science, psychological studies). Documents the shift toward “pay-to-play” API access, jurisdictional clauses that override a user’s home jurisdiction, and concrete misuse cases (e.g., xAI’s Grok generating non-consensual sexual imagery).

Relation to this wiki: the primary methodological anchor for the ### Regulatory & Research Context sections added to every GenGA platform page in Phase 3 of this wiki’s academic-rigor upgrade. Directly studied 5 of this wiki’s 11 GenGA providers (ChatGPT/OpenAI, Claude.ai/Anthropic, DeepSeek, Google Generative AI Services, xAI); citations to the other 6 GenGA providers apply its framework by analogy only.

Pandit, Blankvoort, Shaaban, Luccioni & Birhane (2026)

File: 2603.18964v2.pdf · arXiv:2603.18964v2 [cs.CY] · peer-reviewed, ACM FAccT 2026

“Terms of (Ab)Use: An Analysis of GenAI Services” — manually codes the Terms of Use of six GenAI services (OpenAI/ChatGPT, Anthropic/Claude, Google/Gemini, Microsoft/Copilot, Mistral/Le Chat, DeepSeek) against an EU Unfair Contract Terms Directive lens, finding that all six discard quality/availability assurances, shift output-accuracy responsibility onto users, and use inputs/outputs for purposes beyond the immediate service. Reports inter-annotator disagreement counts (57 cases) as a methodological transparency measure.

Relation to this wiki: the consumer/“abuse” perspective behind the ### Regulatory & Research Context sections on GenGA platform pages. Directly studied 6 of this wiki’s 11 GenGA providers; the other 5 (Llama API, Meta AI, Perplexity, Qwen Chat, xAI) are cited by analogy only in this wiki’s platform pages.

Atkinson (2025)

File: Putting GenAI on Notice_ GenAI Exceptionalism and Contract Law.pdf · Northwestern University Law Review, Vol. 120

“Putting GenAI on Notice: GenAI Exceptionalism and Contract Law” — argues that when a website’s Terms prohibit scraping or AI-training use, and a scraping bot accesses pages containing those Terms, the bot’s deployer has “actual notice” and the prohibition becomes enforceable as a breach-of-contract claim, even though robots.txt itself creates no legal obligation and is “a leaky solution.” Surveys real cases including Perplexity AI’s documented history of ignoring robots.txt.

Relation to this wiki: the legal-enforceability anchor cited throughout scraper_enforcement_gap.md (scraping/training prohibitions embedded in ordinary-looking IP license clauses carry real legal force) and for the ### Legal Context & Research Significance sections on every PGAv2 platform page with a scraping or anti-AI-training clause.

Edwards, Szpotakowski, Cifrodelli, Sangaré & Stewart (2025)

File: private-ordering-generative-ai-and-the-platformisation-paradigm-what-can-we-learn-from-comparative-analysis-of-models-terms-and-conditions.pdf · Cambridge Forum on AI: Law and Governance, Vol. 1, e2 · peer-reviewed, open access

“Private ordering, generative AI and the ‘platformisation paradigm’” — pilot empirical work mapping generative AI providers’ Terms and privacy policies (focused on copyright and data protection), arguing these providers are adopting a “platformisation paradigm”: positioning themselves as “neutral intermediaries” in a manner reminiscent of earlier social-media platforms, which risks repeating past power imbalances between users and platforms if new AI-specific regulation doesn’t account for this private-ordering dynamic.

Relation to this wiki: the structural/rhetorical framing cited throughout the ### Regulatory & Research Context sections of every GenGA platform page — used as a general analytical lens (no specific providers named in the portion of the paper reviewed for this wiki) rather than a provider-specific empirical finding.

Javed & Sajid (2024)

File: 3698393.pdf · ACM Computing Surveys, Vol. 57, No. 2, Article 45 (DOI 10.1145/3698393) · peer-reviewed

“A Systematic Review of Privacy Policy Literature” — systematically reviews 202 peer-reviewed papers on privacy policy analysis published through December 2023, categorizing the field by analysis method (manual, ML/NLP-automated, hybrid; 42.57% of reviewed studies used ML/NLP), by data-protection regulation studied (GDPR, CCPA, etc.), and by sector.

Relation to this wiki: the systematic-review methodology context situating this wiki’s regex/keyword approach relative to the manual-vs-ML spectrum the review documents, cited in the ### Legal Context & Research Significance sections of PGAv2 platform pages.

Soneji, Panda, Neve & Dodge (2025)

File: 2502.08743v1.pdf · arXiv:2502.08743v1 [cs.HC]

“Signed, Sealed,… Confused: Exploring the Understandability and Severity of Policy Documents” — a survey study of user perception of the ToS;DR (Terms of Service; Didn’t Read) clause taxonomy across 243 unique cases, finding clauses are perceived as biased toward the service provider in roughly two-thirds of cases, that users generally find clauses understandable (>72%) yet still disagree sharply on severity (“a polarizing trend”), and that 215 of 243 cases (88%) prompted a user-submitted rewrite.

Relation to this wiki: the user-perception-of-severity finding behind this project’s rule (CLAUDE.md §9.2) of showing only named external critic scores, never a single project-invented severity number — grounding why even a real score is a navigational aid, not a substitute for reading the underlying clause, since even careful human readers disagree about severity.

Note on this wiki’s original citation list

This wiki’s upgrade plan originally named 12 papers, but sources/peer_rev_papers/ contains only the 6 summarized above. The following 6 names were not backed by any file in the folder and are not cited anywhere in this wiki, per an explicit decision to avoid fabricated citations: Soneji et al. (2024) (likely a duplicate/year mix-up with the 2025 paper above, which shares its first author), Zhao et al. (2025), Larrauri et al. (2026), a JMIR (2026) paper, GenAIPABench (2024), and an LLM-based unfair terms detection (2024) paper. If the corresponding PDFs are added to sources/peer_rev_papers/ in the future, this page and the citing pages should be revisited and updated.