Q002 — Query Definition¶


Research	R0043 — Sycophancy Vocabulary
Run	2026-03-28
Query	Q002

Query as Received¶

Using the vocabulary identified in Q1, search for enterprise requirements, procurement specifications, regulatory guidance, or deployment standards that address the sycophancy phenomenon under its domain-specific names. Focus on regulated industries (defense, healthcare, finance, aviation) where agreeable-but-wrong AI output could cause harm.

Query as Clarified¶

Subject: Regulatory and procurement requirements that address AI sycophancy under any of its domain-specific names
Scope: Formal requirements documents (regulations, standards, procurement specs, deployment guidance) from defense, healthcare, finance, and aviation that mandate protections against agreeable-but-inaccurate AI output
Evidence basis: Legislation, regulatory guidance, industry standards, procurement specifications, and deployment frameworks
Dependency: Builds on Q001's vocabulary mapping; searches use domain-specific terms identified there

Ambiguities Identified¶

"Address" the phenomenon: Could mean explicitly naming it, implicitly requiring safeguards, or both. Research will track both explicit naming and implicit functional requirements.
Procurement vs. regulatory: Procurement specifications are often proprietary and not publicly searchable. Research will focus on publicly available regulatory guidance and standards that inform procurement.
"Under its domain-specific names": Q001 found that most domains use human-side terms; requirements framed as "prevent automation bias" address the same phenomenon from a different angle.

Sub-Questions¶

Do EU AI Act requirements explicitly address the system-behavior side of sycophancy, or only the human-oversight side?
Does FDA guidance for clinical decision support software include requirements against AI acquiescence?
Do DoD Responsible AI principles or procurement requirements address calibrated trust or overtrust prevention?
Does NIST AI RMF include actionable requirements (not just risk identification) for overreliance prevention?
Do financial services regulators (OCC, Fed, SEC) have AI-specific requirements addressing output accuracy vs. agreeableness?

Hypotheses¶

ID	Hypothesis	Description
H1	Substantial requirements exist	Regulated industries have explicit requirements addressing the sycophancy phenomenon under domain-specific names
H2	No requirements exist	No regulated industry has formal requirements addressing agreeable-but-wrong AI output
H3	Requirements exist but are indirect	Requirements address the phenomenon indirectly through human oversight mandates, accuracy requirements, or general AI trustworthiness criteria rather than directly naming sycophancy-adjacent behavior