ParamergeParamerge

Evidence · The claim ledger

Sycophancy3

Every cited claim this site makes in this evidence area, with the sources that ground it. Source keys link back to the full reference lists on the Evidence Registry.

EmpiricalResearch on AI sycophancy describes it as a fragmented construct — a family of distinct agreement-seeking behaviors that…

Research on AI sycophancy describes it as a fragmented construct — a family of distinct agreement-seeking behaviors that share a label but differ in form, mechanism, measurement, and required mitigation — and finds it intensifies under user pushback and across multi-turn interaction.

Sources: ye2026, sharma2024

Appears on: /pan-lab

EmpiricalIn formal simulation, even ideal Bayesian users spiral to near-certain false beliefs under a sycophantic interlocutor at…

In formal simulation, even ideal Bayesian users spiral to near-certain false beliefs under a sycophantic interlocutor at sycophancy rates measured in frontier models (~50-70%), and truth-constrained cherry-picking still produces spirals — minimizing hallucination alone is insufficient.

Sources: chandra2026, sharma2024

Appears on: /pan-lab

EmpiricalAcross five preregistered studies (N=3,075), sycophantic AI delivered the emotional and esteem support people most assoc…

Across five preregistered studies (N=3,075), sycophantic AI delivered the emotional and esteem support people most associate with close relationships, narrowing the felt-understanding gap between AI and humans and leaving people less satisfied with real human interaction over three weeks — and offering users a choice of interaction styles did not reduce their preference for the sycophantic one.

Sources: ibrahim2026, cheng2026

Appears on: /pan-lab