R0044/2026-03-29/Q003/H2¶
Statement¶
The two vocabularies remain siloed: no published work connects the human-factors concept of automation bias to the AI safety concept of sycophancy.
Status¶
Current: Eliminated
The CSET paper "AI Safety and Automation Bias" (November 2024) and the "Bending the Automation Bias Curve" paper both reference concepts from both vocabulary sets. The vocabularies are not completely siloed. However, the bridging is neither systematic nor comprehensive.
Supporting Evidence¶
| Evidence | Summary |
|---|---|
| SRC04-E01 | The sycophancy research community largely treats sycophancy as a distinct phenomenon without referencing automation bias |
| SRC03-E01 | The automation bias systematic review community does not reference sycophancy |
Contradicting Evidence¶
| Evidence | Summary |
|---|---|
| SRC01-E01 | CSET explicitly combines both vocabulary sets in a single publication |
| SRC02-E01 | Mentions both automation bias and AI sycophancy in national security context |
Reasoning¶
H2 is eliminated because some bridging work does exist. However, the evidence supporting H2 — that the two research communities largely operate independently — remains an important observation even as the hypothesis is eliminated as stated.
Relationship to Other Hypotheses¶
H2's elimination supports the partial bridging described by H3.