AI-generated ethics exercise materials carried external citations that did not survive checking, ranging from misattributed to apparently fabricated¶

Claim · All claims
Evidence · 1 study · design r2 · q1
Currency · newest study 2026 · not yet checked against its sources · not reviewed by a person · recheck by 2029: tool and model accuracy changes quickly

Subclaims¶

q1 i? All three external citations in the Gemini-generated ethics materials failed verification: one misattributed a claim to Chan and Hu, one attached a nonexistent framework to real 2024 OSPI guidance, and one could not be verified at all. → Yusuf Pisan 2026

Evidence¶

Yusuf Pisan 2026¶

Yusuf Pisan. (2026). Teaching Intro AI When the Tools Can Do the Homework: A Course Redesign and a Student Bill of Rights. https://arxiv.org/abs/2608.05175 doi:10.48550/arxiv.2608.05175

q1 · i? · design · r2

The instructor's post-hoc verification of the three citations embedded in the Gemini-generated ethics exercises. He reports that a survey by Chan and Hu was cited for a finding it does not make, that a Washington OSPI H-AI-H framework does not exist in the actual guidance, and that a Cornell item could not be found.

"None of the three reproduces as cited. I treat this as an exhibit rather than an aside: a tool used to generate ethics materials about AI inserted references that range from misattributed to apparently fabricated, inside an exercise whose own subject was the reliability of AI in teaching."

Discussion¶

Related Claims¶

• AI output errors are frequent: 44.5% fabricated references in drafted research proposals and 5–13% unsafe chatbot medical answers — a broader claim this one bears on
• The most frequently cited ethical concerns about GenAI in research were hallucinated facts, fabricated references, and misleading content — related
• Review-attributed evidence: fabricated model-generated references clustered in disciplinary journals in a student verification exercise — related
• Of 30 verified hallucinated references, 13 appeared entirely fabricated and 17 were hybrid references combining a real title with fabricated authorship or metadata — a broader claim this one bears on
• Conservative manual verification identified 30 hallucinated references across 14 papers in 2025 and 2026 computing education venues — related
• AI-assisted scientific writing assignments risk fabricated references and altered chemical meaning — possibly the same claim (merge candidate)
• ChatGPT fabricates references: the authors caught it inventing a non-existent 'Ribeiro and Vala, 2020' article with full bibliographic details — related
• Checking can coexist with unresolved errors: 7 of 15 students failed a task containing plausible LLM falsehoods despite self-checking — related
• Experimental-group students detected AI errors and enacted verification and ethical self-regulation as routine practices — related
• Most of the article's body text was generated by ChatGPT from author prompts, and all references ChatGPT inserted were fictional — related
• Evidence types differ systematically in probative value: detection outputs cluster at weak inferential force while admissions, observed exam behaviour and fabricated references carry the strongest profiles — related