People adopted false facts from confident AI tutors
Source check: Source checked. Not independent scientific confirmation.
Finding: In a study with 32 volunteers, participants selected an assistant’s planted false answer for about 60% of quiz items assigned false information, compared with 11% of items assigned correct information. Mean scores fell from 7.4 to 2.9 out of 10 between fully accurate and fully falsified GPT-5-mini briefings.
Technology: GPT-5-mini
Family GPT-5: profile · models page
Publication: Preprint · Test context: direct test. Catalog categories, not an evidence grade.
- AI contribution
- GPT-5-mini chat assistants explained science topics from briefings containing controlled levels of plausible false claims.
- Test and result
- The counterbalanced study used five briefing reliability levels and 104 completed quizzes, with independently randomized topic order and corrective debriefing.
- Limitation / still unknown
- This was a small study with one underlying model, deliberately confident responses, and immediate quizzes. Retention over longer periods and responses to other assistants were not assessed. Participants could continue consulting the same assistant during each quiz. The measured choices do not establish unaided memory retention or durable internalized beliefs.
Evidence sources and annotation context
Evidence context (descriptions follow the cited sources; test type is a catalog classification):
Evidence sources: https://arxiv.org/abs/2610.31579 · https://arxiv.org/html/2610.31579v1
Source checked · editorial details
The source page was checked against this entry. Independent confirmation is not assessed by this label.
Source type: paper. Paper: linked separately.
Review date: 21 Nov 2026. Review scope: arXiv v1 trial design §§3.1–3.7, Table 3, §4.2 and ethics statement.
No follow-up recorded; this is not a claim that no follow-up exists. Follow-up review scope and date are not recorded. No source-supported relationships recorded.
Editorial assessment: Impact 2/5 · Validation 3/5 · Novelty 3/5. Impact and novelty are editorial judgments; validation follows the annotated test type where one exists. Not an evidence grade. Scoring criteria.