Skip to content

R0043/2026-03-28/Q003/H1

Research R0043 — Sycophancy Vocabulary
Run 2026-03-28
Query Q003
Hypothesis H1

Statement

Multiple researchers and organizations have identified the AI safety / regulated-industry vocabulary gap and are actively working to bridge it with shared taxonomies.

Status

Current: Partially supported

The broader terminology gap is well-recognized, and multiple organizations are building risk taxonomies and glossaries. However, these efforts are general — they do not specifically target the sycophancy/overreliance vocabulary gap.

Supporting Evidence

Evidence Summary
SRC01-E01 Trilateral Research explicitly identifies "the AI terminology gap" as an operational problem
SRC02-E01 Standardized Threat Taxonomy explicitly addresses the "Tower of Babel problem" between engineering and legal teams
SRC03-E01 MIT AI Risk Repository synthesizes 65 documents into shared risk framework
SRC05-E01 AIR 2024 maps 314 risk categories across government and corporate policies

Contradicting Evidence

Evidence Summary
SRC04-E01 Argues the problem is deeper than taxonomy — existing concepts do not translate across the human/machine divide

Reasoning

H1 is partially supported because significant taxonomy-building efforts exist (MIT, AIR 2024, threat taxonomy, Trilateral Research). However, none of these specifically address the sycophancy/overreliance vocabulary gap. They address the broader challenge of AI risk terminology, within which sycophancy is a small component.

Relationship to Other Hypotheses

The evidence pattern matches H3 more closely than H1: the gap is recognized broadly but not for sycophancy specifically.