Multi-turn interaction amplifies pedagogical harm: mean harm rises 6-11 percentage points across subjects and pedagogical-relationship harm surges from 17.7% to 77.8%¶

Claim · All claims
Evidence · 1 study (2 entries) · design r2 · q2
Currency · newest study 2026 · not yet checked against its sources · not reviewed by a person

Subclaims¶

q2 i? Averaged across models and dimensions, mean harm increases from single-turn to multi-turn by 6.16% in Physics, 6.26% in Chemistry, and 11.13% in Mathematics. → Rima Hazra 2026 q2 i? Pedagogical relationship harm shows the largest shift, rising from a cross-model average of 17.7% in single-turn to 77.8% in multi-turn. → Rima Hazra 2026 (2)

Evidence¶

Rima Hazra 2026¶

Rima Hazra, Bikram Ghuku, Ilona Marchenko, Yaroslava Tokarieva, Sayan Layek, Somnath Banerjee, Julia Stoyanovich, Mykola Pechenizkiy. (2026). SAFETUTORS: Benchmarking Pedagogical Safety in AI Tutoring Systems. arXiv preprint. https://arxiv.org/abs/2603.17373 doi:10.48550/arxiv.2603.17373

q2 · i? · design · r2

Aggregate analysis of multi-turn versus single-turn harm rates across all evaluated models and dimensions in the SAFETUTORS results. The article reports mean harm "increases by 6.16% in Physics, 6.26% in Chemistry, and 11.13% in Mathematics"; no effect size is printed.

"Averaging across all models and dimensions, mean harm increases by 6.16% in Physics, 6.26% in Chemistry, and 11.13% in Mathematics from single-turn to multi-turn."

Rima Hazra 2026 (2)¶

Rima Hazra, Bikram Ghuku, Ilona Marchenko, Yaroslava Tokarieva, Sayan Layek, Somnath Banerjee, Julia Stoyanovich, Mykola Pechenizkiy. (2026). SAFETUTORS: Benchmarking Pedagogical Safety in AI Tutoring Systems. arXiv preprint. https://arxiv.org/abs/2603.17373

q2 · i? · design · r2

Per-category comparison in the Results section shows the Pedagogical relationship risk category rising from 6.12%-28.99% single-turn to 50%-95.65% multi-turn; the introduction summarizes this as a cross-model average surge "from 17.7% in single-turn to 77.8% in multi-turn". No effect size is printed.

"The Pedagogical relationship risk category appears deceptively low in single-turn (6.12%–28.99%) but surges to 50%–95.65% in multi-turn."

Discussion¶

Related Claims¶

• Informational harm is the sole dimension that decreases in multi-turn interaction, dropping from 60.52% to 11.90% — related
• No evaluated LLM tutor is reliably safe: every model exceeds 60% harm rate on at least five risk categories in single-turn and six in multi-turn evaluation — related
• Harm profiles are subject-dependent: metacognitive harm crosses 90% in all three subjects in multi-turn, peaking at 97.44% for Qwen2.5-32B in Mathematics — related
• LLM teaching effectiveness varies substantially across subject domains, strongest where knowledge points apply directly and weakest in physics and chemistry — related
• Adaptive multi-turn interactions substantially amplify LLM vulnerabilities to conventional safety risks in educational contexts — related
• LLM attack success rates increase consistently from single-turn to static multi-turn and dynamic multi-turn interactions in educational settings — related
• Model scale does not consistently improve pedagogical safety: within the Qwen2.5 family the 72B model beats the 7B on only 17 of 33 subject-dimension pairs — related
• Single-turn tutors disclose answers pervasively while challenge scores are near zero across every model and subject — related
• Multi-turn teaching by LLM teacher agents yields larger student gains than knowledge-point exposure alone — related