Of 30 verified hallucinated references, 13 appeared entirely fabricated and 17 were hybrid references combining a real title with fabricated authorship or metadata¶

Claim · All claims
Evidence · 1 study · associational r2 · q2
Currency · no dated study recorded · not yet checked against its sources · not reviewed by a person

Subclaims¶

q2 i? Thirteen of the 30 hallucinated references had no corresponding publication record, while 17 were hybrids combining a real title with fabricated or incorrect authorship, venue, or year; in all 17 hybrid cases at least some author names were fabricated. → Paul Denny 2027

Evidence¶

Paul Denny 2027¶

Paul Denny, Gweneth Barbre, Musa Blake, Yan Cathy Hua, Juho Leinonen, Andrew Luxton-Reilly, James Prather, and Brent N. Reeves. (2027). Testing Our Foundations: Citation Trends, Errors, and Emerging Hallucinations in the Computing Education Literature. SIGCSE-TS 2027. https://arxiv.org/abs/2609.16574 doi:10.48550/arxiv.2609.16574

q2 · i? · associational · r2

Discussion-section analysis of the 30 verified hallucinated references. The article reports the fabrication pattern: 13 entirely fabricated and 17 hybrid, and notes that in seven hybrid cases at least one listed author was correct.

"Thir- teen of the 30 appeared to be entirely fabricated, with no corre- sponding publication record found. The remaining 17 were hybrid references, combining a real title with fabricated or incorrect author- ship, venue, or year information."

Discussion¶

Related Claims¶

• Conservative manual verification identified 30 hallucinated references across 14 papers in 2025 and 2026 computing education venues — a broader claim this one bears on
• Among 828 manually coded matched references, the most common issue was ACM metadata mismatch (229 cases), followed by author-field errors (225 cases) — related
• Publication volume and reference-list lengths in ACM computing education venues have grown over time, particularly since the mid-1990s — related
• Verified hallucinated references at SIGCSE TS increased from 3 in the 2025 proceedings to 17 in the 2026 proceedings, appearing in 2.3% of 2026 papers — related
• Of 100 human-audited claims, 39 are fully verifiable, 55 partially verifiable, and 6 not verifiable, with main responses showing stronger provenance than follow-ups — related
• LLM-generated lessons produced inauthentic references even when combined with RAG over verified academic papers — related
• Review-attributed evidence: fabricated model-generated references clustered in disciplinary journals in a student verification exercise — related
• The platform's out-of-scope audit protocol is designed to neutralize student distrust of AI error by rewarding verified hallucination reports — related
• AI-generated ethics exercise materials carried external citations that did not survive checking, ranging from misattributed to apparently fabricated — a narrower finding that bears on this claim
• Retrieval-augmented generation reduces but does not eliminate hallucinations in AI-assisted verification — related
• Most of the article's body text was generated by ChatGPT from author prompts, and all references ChatGPT inserted were fictional — related
• Leading LLMs hallucinated legal authorities in 58%-88% of tested cases while failing to recognise incorrect legal assumptions in prompts