Skip to content

R0043/2026-04-01/Q003/SRC01/E01
Research R0043 — Sycophancy Vocabulary
Run 2026-04-01
Query Q003
Source SRC01
Evidence SRC01-E01
Type Analytical

Trilateral Research diagnosis of the AI terminology gap and proposed solutions

URL: https://trilateralresearch.com/responsible-ai/how-to-fix-the-ai-terminology-gap

Extract

Core diagnosis: "Ask five experts to define 'AI regulation' or 'AI risk' and you may hear ten different answers." The problem extends to cross-domain misalignment:

  • "Transparency obligations" (policy) maps to "data provenance disclosures + model cards" (engineering)
  • "Right to explanation" (legal) maps to "feature attribution methods like SHAP" (technical)

Five proposed solutions:

  1. Minimal glossary (10-20 core terms) — Define: explainability, robustness, bias/fairness, safety, monitoring, red-teaming
  2. Risk taxonomy — Organize risks by origin: training data, inference, output, non-technical, agentic behaviors
  3. Translation tables — Map legal language to implementable technical controls with evidence requirements
  4. Sociotech integration — Involve DPOs and cybersecurity officers early in AI lifecycles
  5. Operational embedding — Revise DPIAs, model cards, threat models, and supplier questionnaires to reference the glossary

Notable gap: The proposed minimal glossary does NOT include sycophancy, agreeableness, or related behavioral terms. The focus is on high-level governance concepts.

Relevance to Hypotheses

Hypothesis Relationship Strength
H1 Supports Vocabulary gap explicitly identified as a problem with proposed solutions
H2 Contradicts The gap IS recognized
H3 Supports Solutions focus on governance terms, not behavioral risks like sycophancy

Context

Trilateral Research is a consultancy that provides responsible AI services, which creates a potential incentive to emphasize the terminology gap. However, their diagnosis is consistent with independent academic findings and their solutions are concrete and actionable.