Skip to content

R0041/2026-03-28/Q002 — ACH Matrix

Matrix

H1: Explicit sycophancy requirements exist H2: Problem not on radar H3: Concern exists, different framing
SRC01-E01: Georgetown CSET "caving to expectations" + -- ++
SRC02-E01: Mass General 100% sycophancy failure + -- ++
SRC03-E01: Science study cross-sector risk + -- +
SRC04-E01: Georgetown policy "open questions" - - ++
SRC05-E01: FAA no sycophancy mention -- + +
SRC06-E01: FINRA hallucination/bias not sycophancy -- + ++

Legend: - ++ Strongly supports - + Supports - -- Strongly contradicts - - Contradicts - N/A Not applicable to this hypothesis

Diagnosticity Analysis

Most Diagnostic Evidence

Evidence ID Why Diagnostic
SRC05-E01 FAA's omission of sycophancy in its AI safety framework strongly discriminates between H1 (which predicts it would be there) and H3 (which explains why it is not)
SRC04-E01 Georgetown framing sycophancy standards as "open questions" directly indicates these requirements do not yet exist

Least Diagnostic Evidence

Evidence ID Why Non-Diagnostic
SRC03-E01 The Science study documents the problem across sectors but does not discriminate between whether sectors have formalized requirements

Outcome

Hypothesis supported: H3 — The concern exists across sectors but is framed using domain-specific terminology and has not been formalized as a discrete "sycophancy reduction" requirement.

Hypotheses eliminated: H2 — The problem is clearly on the radar of defense, healthcare, and policy researchers.

Hypotheses inconclusive: H1 — Partially supported by academic recommendations but not by formal requirements or procurement specifications.