Claims

AI output errors are frequent: 44.5% fabricated references in drafted research proposals and 5–13% unsafe chatbot medical answers

Claim · All claims
Evidence · 1 study · theoretical r3 · q2
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

Reading the evidence codes

q is the design tier, i the impact, n the sample, and the kind with its r the rigour judged within that kind: what each code means.

This claim: no standardized effect is recorded in observations/ for it yet, so no benchmark reading is possible.

q2 i? A check of research proposals drafted by a frontier model found 44.5% of references fabricated, and a physician audit judged 5–13% of public chatbot medical answers outright unsafe. → Marcus Kubsch 2026

Evidence

Marcus Kubsch 2026

Marcus Kubsch. (2026). AI as a Partner in Learning about, Doing, and Engaging with Science: Vigilance as the Key to Productive Augmentation. https://arxiv.org/abs/2606.16822 doi:10.48550/arxiv.2606.16822

q2 · i? · theoretical · r3

The paper cites two audits as evidence that AI errors exist in research and everyday settings: the review reports, citing Ghareeb et al. (2026), that 44.5% of references in frontier-model-drafted proposals were fabricated, and, citing Draelos et al. (2026), that 5–13% of public chatbot medical answers were judged outright unsafe. Second-hand; designs not described.

"a check of research proposals drafted by a frontier model without grounding in the literature found 44.5% of their references fabricated (Ghareeb et al., 2026), and a physician audit of the public chatbots people actually consult judged 5–13% of their answers to patient-posed medical questions outright unsafe (Draelos et al., 2026)."

Discussion

Page metadata
Field Value
type claim
id ai-output-error-rates-fabricated-unsafe
title AI output errors are frequent: 44.5% fabricated references in drafted research proposals and 5–13% unsafe chatbot medical answers
description AI output errors are frequent: 44.5% fabricated references in drafted research proposals and 5–13% unsafe chatbot medical answers
status review
evidence_strength weak
newest_evidence 2026
recheck_by 2029
generated by process:wiki-ingest on 2026-10-10
trust unverified — no one has confirmed this page's content
sources 1 — marcus-kubsch-2026

Frontmatter is the page's machine-readable record — see the Schema & Guide.