Research map / Large language models / AI and society

Stop DDoS Attacking the Research Community with AI-Generated Survey Papers

NeurIPS 2025 · Large language models › AI and society · approach: Critiques of LLM claims and judging · 2 citations (OpenAlex)

Read the paperCheck your idea against 108,967 papers

AI-extracted summary. Written by AI from the paper's full text. Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Position paper arguing that the flood of AI-generated survey papers is a "survey paper DDoS attack" on the research community, supported by arXiv trend analysis and AI-detector scores, and proposing policy norms and "Dynamic Live Surveys".

Problem

The unchecked proliferation of low-quality, redundant, or hallucinated AI-generated survey papers on preprint servers such as arXiv, which overwhelms researchers, burdens reviewers, and erodes trust in the scientific record.

Approach

Frames AI-generated survey flooding as a "survey paper DDoS attack" and argues for norms (disclosure, stricter survey review, redundancy limits, AI-detection screening) plus a community-maintained, version-controlled "Dynamic Live Surveys" infrastructure combining LLM ingestion agents with expert curation.

Key results

• Post-2022 spike in AI-generated scores for CS survey papers on arXiv, corroborated by DeTeCtive and MAGE detectors (e.g., MAGE 2022-2023 growth ratio 0.7058).
• Citation overlap between post-2022 surveys exceeds 60% (Jaccard >0.5) versus below 40% (Jaccard <0.3) pre-2022.
• Average embedding similarity of survey abstracts rose from 0.6033 in 2022 to 0.8367 in 2023 and 0.7986 in 2024.

Surprising findings

• Some survey papers appear under anonymous groups or collectives with no clear institutional link, with single authors or small teams submitting several surveys in a short period.

Limitations

• Trend analysis relies on keyword-based survey identification and open-source AI-content detectors whose false positives are acknowledged but not quantified.
• Citation overlap and quality claims are based on small, unspecified samples (e.g., 10 surveys on an ML topic) and are described as not definitive proof.
• Proposed Dynamic Live Surveys framework is conceptual and not implemented or evaluated.

Experimental setup

Datasets: arXiv CS survey submissions 2020-2024 (10,063 papers)

Metrics: AI-generated content score citation overlap percentage Jaccard index embedding cosine similarity

Builds on: SurveyForge

Closest papers by meaning

• Consent in Crisis: The Rapid Decline of the AI Data Commons (NeurIPS 2024 · similarity 0.83)
• Monitoring AI-Modified Content at Scale: A Case Study on the Impact of ChatGPT on AI Conference Peer Reviews (ICML 2024 · similarity 0.81)
• DeepTRACE: Auditing Deep Research AI Systems for Tracking Reliability Across Citations and Evidence (ICLR 2026 · similarity 0.81)
• Position: A Safe Harbor for AI Evaluation and Red Teaming (ICML 2024 · similarity 0.81)
• The Unreasonable Effectiveness of Open Science in AI: A Replication Study (AAAI 2025 · similarity 0.81)
• The Stronger the Diffusion Model, the Easier the Backdoor: Data Poisoning to Induce Copyright BreachesWithout Adjusting Finetuning Pipeline (ICML 2024 · similarity 0.81)