Skip to content

R0043/2026-04-01/Q003/H1

Research R0043 — Sycophancy Vocabulary
Run 2026-04-01
Query Q003
Hypothesis H1

Statement

The vocabulary gap between AI safety and regulated-industry terminology has been explicitly identified as a problem, and there are active efforts to create shared taxonomies.

Status

Current: Supported

Supporting Evidence

Evidence Summary
SRC01-E01 Trilateral Research (2025) explicitly diagnoses the gap and proposes solutions
SRC02-E01 CSIRO/UNSW harmonised terminology framework for AI evaluation
SRC03-E01 Huwyler's threat taxonomy bridges technical and business language

Contradicting Evidence

Evidence Summary
SRC04-E01 IAPP glossary (100+ terms) does not include sycophancy, suggesting the specific sycophancy vocabulary gap is not yet recognized as a priority

Reasoning

Multiple independent sources confirm the AI terminology gap is recognized as a problem. However, the specific sycophancy vocabulary gap (as opposed to the broader AI governance terminology gap) has received less attention. Most efforts focus on harmonizing high-level concepts (fairness, transparency, robustness) rather than specific behavioral risks like sycophancy.

Relationship to Other Hypotheses

H1 is supported at the general level but qualified by H3 — the sycophancy-specific gap is a subset of the broader recognized problem. H2 is eliminated.