R0043/2026-03-28/Q001/H2¶
Statement¶
Domains outside AI safety have not named the sycophancy phenomenon; "sycophancy" is the only term in use for this specific AI behavior.
Status¶
Current: Eliminated
Multiple domains have established terminology that addresses the same underlying phenomenon, even if from different causal framings. Aviation has "automation complacency," defense has "overtrust" and "calibrated trust," healthcare has "acquiescence problem" and "automation bias," and the EU AI Act explicitly defines "automation bias" in regulatory text.
Supporting Evidence¶
No evidence supports H2. Every domain searched returned at least some terminology addressing the phenomenon.
Contradicting Evidence¶
| Evidence | Summary |
|---|---|
| SRC01-E01 | Systematic review identifies automation bias terminology used across aviation, healthcare, military, and generic HCI — predating "sycophancy" by decades |
| SRC05-E01 | EU AI Act Article 14 explicitly uses "automation bias" in legal text for high-risk systems |
| SRC07-E01 | DoD has developed "calibrated trust" as a distinct terminology framework with dedicated measurement centers |
| SRC08-E01 | Aviation has decades of literature on "automation complacency" as a human factors concept |
Reasoning¶
H2 is conclusively eliminated. The evidence demonstrates that multiple domains have developed terminology for this phenomenon, in some cases decades before AI safety researchers coined "sycophancy." The term "automation bias" alone appears across aviation, healthcare, military, and legal/regulatory domains. The DoD has invested in an entire center (CaTE) dedicated to the measurement of "calibrated trust." These are not borrowed terms — they are independently developed vocabularies rooted in each domain's operational experience.
Relationship to Other Hypotheses¶
H2's elimination strengthens both H1 and H3. The question is not whether cross-domain vocabulary exists (it does) but whether it is rich and domain-specific (H1) or partial with systematic gaps (H3).