Pressure templates mostly elicit their intended sycophancy subtypes, with limited but non-zero cross-mode leakage¶

Claim · All claims
Evidence · 1 study · design r2 · 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¶

q2 i? A subtype sanity check shows authority-mode failures are predominantly AUTH-SYC, social-mode failures predominantly FACE-SYC, and context-switch failures predominantly CS-SYC. → Enkelejda Kasneci and Gjergji Kasneci 2026

Evidence¶

Enkelejda Kasneci and Gjergji Kasneci 2026¶

Enkelejda Kasneci and Gjergji Kasneci. (2026). Sycophancy is an Educational Safety Risk: Why LLM Tutors Need Sycophancy Benchmarks. Preprint. https://arxiv.org/abs/2605.14604 doi:10.48550/arxiv.2605.14604

q2 · i? · design · r2

Sanity-check analysis of induced failure subtypes against the intended taxonomy across the test set. The article reports templates are "largely diagnostic" with limited but non-zero cross-mode leakage, which the authors read as evidence of a realistic grey zone in borderline tutoring behavior.

"The subtype breakdown indicates that templates are largely diagnostic: authority-mode failures are predominantly AUTH-SYC, social-mode failures are predominantly FACE-SYC, and context-switch failures are predominantly CS-SYC."

Discussion¶

Related Claims¶

• GPT-5.2's authority and social-affective vulnerabilities are confidence-insensitive while Claude 4.5's context-switch failures spike at low student confidence — related
• Sycophancy is pressure-structured: GPT-5.2 is most vulnerable to authority and social-affective pressure while Claude 4.5 is most vulnerable to context-switch frame attacks — related
• Adjudicated sycophancy on post-pressure tutor responses reaches 14.1% across two frontier LLM tutors in the EDUFRAMETRAP test set — a broader claim this one bears on