Phantom References: Hallucinated Citations That Survive Peer Review at Top-Tier Conferences

Summary

Measures identity-level citation hallucination (non-existent works, substantial author-list mismatches) in accepted camera-ready ICLR, ICML, NeurIPS and USENIX Security papers using RefChecker, an open-sourced multi-source verification pipeline. Reference-level rates stay under 1%, but roughly one in twenty 2025 NeurIPS and USENIX Security papers carries at least two likely hallucinated references, with post-ChatGPT increases and hallucinations even in award-winning papers. Shows peer review does not enforce citation integrity while auditing costs about $0.04 per paper — a clean argument for automated integrity infrastructure.

Contribution

Queued — not yet read.

Method

Queued — not yet read.

Relevance to RISE

Surfaced by the 2026-08-29 literature sweep as a high-priority candidate for the RISE knowledge base (angle: AI in peer review). Verified against: arXiv abstract page opened 2026-08-29 (title, 3 authors, submitted 1 Jul 2026, full abstract read)

Critique / open questions

Queued — not yet assessed.

Key quotes

Large language models can generate polished scientific text that includes unsupported claims, allowing hallucinations to enter the archival record. (abstract)

Assessing this risk via technical statements is difficult and often requires expert judgment, but citations provide a more auditable surface: a reference either resolves to a real scholarly work with compatible authorship, or it does not. (abstract)