Skip to content

R0043/2026-04-01/Q003/H3

Research R0043 — Sycophancy Vocabulary
Run 2026-04-01
Query Q003
Hypothesis H3

Statement

The broader AI terminology gap has been recognized, but the specific sycophancy vocabulary gap is a smaller, not-yet-prioritized subset of the larger problem. Existing efforts focus on high-level governance concepts, not specific behavioral risks.

Status

Current: Supported (best fit)

Supporting Evidence

Evidence Summary
SRC04-E01 IAPP glossary covers 100+ terms but excludes sycophancy and related behavioral safety terms
SRC01-E01 Trilateral's proposed minimal glossary focuses on high-level concepts (explainability, robustness, fairness) not behavioral risks
SRC02-E01 CSIRO framework focuses on evaluation methodology terms, not behavioral AI risks

Contradicting Evidence

Evidence Summary
SRC03-E01 Huwyler's threat taxonomy is operationally detailed (53 sub-threats) but still does not include sycophancy

Reasoning

The evidence consistently shows that the broader AI terminology gap is well-recognized and actively being addressed. However, current efforts focus on governance-level concepts (fairness, transparency, explainability) and technical security threats (adversarial attacks, prompt injection), not on specific behavioral risks like sycophancy. The sycophancy vocabulary gap falls into the space between high-level governance and low-level security — behavioral model properties that affect safety but do not fit neatly into either category.

Relationship to Other Hypotheses

H3 is the most accurate characterization. It subsumes H1 (gap recognized) while adding the qualification that sycophancy-specific attention is lacking. It explains why the gap persists despite recognition of the broader problem.