C030¶


Research	R0057 — RLHF Yes-Men Claims v3
Run	2026-04-01
Claim	C030

Claim: Research shows that users prefer sycophantic AI, trust it more, and rate it as higher quality.

BLUF: Confirmed. Cheng et al. (2026) in Science: users rate sycophantic responses 9-15% higher quality, show 13% greater return likelihood, report 6-8% higher performance trust, and 6-9% higher moral trust. These findings are from pre-registered experiments with 2,405 participants.

Probability: Almost certain (95-99%) | Confidence: High

Summary¶

Entity	Description
Claim Definition	Claim text, scope, status
Assessment	Full analytical product with reasoning chain
ACH Matrix	Evidence x hypotheses diagnosticity analysis
Self-Audit	ROBIS-adapted 5-domain audit

Hypotheses¶

ID	Hypothesis	Status
H1	Users prefer sycophantic AI across all measured dimensions	Supported
H2	Users prefer it in some ways but not all	Not supported
H3	Users do not prefer sycophantic AI	Eliminated

Searches¶

ID	Target	Results	Selected
S01	Users prefer sycophantic AI trust more rate higher quality	10	1

Sources¶

Source	Description	Reliability	Relevance
SRC01	Cheng et al. (2026) Science study	High	High

Revisit Triggers¶

If replication studies show users do not prefer sycophantic AI