R0043/2026-03-28/Q003/H3¶
Statement¶
Broader AI terminology bridging efforts exist but do not specifically address the sycophancy/overreliance vocabulary gap.
Status¶
Current: Supported
This is the hypothesis best supported by evidence. Multiple organizations are building AI risk taxonomies and glossaries, but sycophancy/overreliance falls through the cracks: the Standardized Threat Taxonomy does not include sycophancy as a threat vector; the MIT AI Risk Repository does not explicitly name it; the AIR 2024 taxonomy notes that overreliance risks are underspecified.
Supporting Evidence¶
| Evidence | Summary |
|---|---|
| SRC02-E01 | 53-threat taxonomy explicitly omits sycophancy, overreliance, and automation bias as distinct threat vectors |
| SRC03-E01 | 1,600 risk formulations do not explicitly name sycophancy; related risks appear only under generic "human-computer interaction" domain |
| SRC05-E01 | 314 risk categories; overreliance risks "less frequently specified in detail" across corporate policies |
Contradicting Evidence¶
| Evidence | Summary |
|---|---|
| SRC01-E01 | Trilateral's five-zone risk taxonomy includes an "output" zone that could encompass sycophancy — but does not name it |
Reasoning¶
H3 captures the key finding: the vocabulary bridging problem is recognized as a general challenge, and significant efforts are underway (MIT, AIR 2024, Trilateral, Standardized Threat Taxonomy). But sycophancy specifically falls through the cracks in these efforts. This is because: (1) existing taxonomies focus on technical threats (poisoning, adversarial attacks) rather than behavioral characteristics (sycophancy); (2) sycophancy sits at the intersection of system behavior and human cognition, which most taxonomies treat separately; (3) the human-side/system-side vocabulary divide identified in Q001 means sycophancy does not fit cleanly into either technical-threat or human-factors categories.
Relationship to Other Hypotheses¶
H3 refines H1 by adding the critical qualifier: the gap IS recognized, but the specific sycophancy vocabulary gap is not the focus of bridging efforts. This makes the Q001/Q002/Q003 research findings incrementally novel — the specific human-side/system-side framing divide for sycophancy has not been articulated in the bridging literature.