Q002 — Query Definition¶


Research	R0048 — Corporate AI Training
Run	2026-04-01
Query	Q002

Query as Received¶

Do any corporate or government AI training materials specifically warn users about sycophancy — the tendency of AI to agree with the user, provide comforting answers without evidence, confirm user assumptions rather than challenge them, or prioritize helpfulness over accuracy? Search using both the AI safety term "sycophancy" and the human-factors equivalents: automation bias, overtrust, overreliance, confirmation reinforcement, acquiescence. Include any training that warns users the AI may tell them what they want to hear rather than what is true.

Query as Clarified¶

This query asks whether the specific concept of AI sycophancy (or functionally equivalent concepts under different terminology) appears in corporate or government training materials. The query correctly identifies that sycophancy may be described differently across domains:

AI safety/ML: sycophancy, reward hacking, preference alignment failure
Human factors: automation bias, overtrust, overreliance, complacency
Psychology: confirmation bias reinforcement, acquiescence bias
Plain language: "tells you what you want to hear," "yes-man AI"

The query contains an embedded expectation (from the researcher profile): "Expectation that sycophancy is not covered in training materials." This must be tested, not assumed.

Sub-questions: 1. Does any training material use the word "sycophancy" in the context of AI behavior? 2. Does any training material warn about the specific behavior (AI agreeing/confirming) even if using different terminology? 3. Does any training address the human-factors equivalents (automation bias, overtrust, overreliance)? 4. Is there any training that warns users the AI may tell them what they want to hear?

BLUF¶

No publicly visible corporate or government AI training program uses the term "sycophancy" or explicitly warns employees that AI systems may tell them what they want to hear. The concept appears extensively in AI safety research, policy analysis (Georgetown, Brookings), and academic literature (Science journal), but has not yet propagated into employee training materials. The NHS framework addresses the adjacent concept of automation bias, and the Lumenova/Microsoft analysis documents scenarios "where AI might falter," but neither connects these to the specific behavioral pattern of AI prioritizing user agreement over accuracy. There is a complete disconnect between the AI safety community's understanding of sycophancy and what employees are taught.

Scope¶

Domain: AI sycophancy, automation bias, overtrust in corporate/government training
Timeframe: 2023-2026
Testability: Verifiable through publicly available training materials, policy documents, and expert recommendations

Assessment Summary¶

Probability: N/A (open-ended query)

Confidence: Medium-High

Hypothesis outcome: H2 (sycophancy is absent from training but adjacent concepts partially addressed) is best supported.

[Full assessment in assessment.md.]

Status¶

Field	Value
Date created	2026-04-01
Date completed	2026-04-01
Researcher profile	Phillip Moore
Prompt version	Unified Research Methodology v1
Revisit by	2026-10-01
Revisit trigger	Any major training provider adds sycophancy to curriculum; NIST or ISO publish AI sycophancy testing standards; Georgetown/Brookings policy recommendations adopted by regulators