R0044/2026-04-01/Q001/H2¶
Statement¶
Requirements exist in partial or emerging form — some frameworks identify the problem of AI systems reinforcing user assumptions and recommend system-level mitigations, but these are not yet specific, enforceable mandates constraining AI output behavior.
Status¶
Current: Supported
Supporting Evidence¶
| Evidence | Summary |
|---|---|
| SRC02-E01 | EU AI Act Article 14 requires systems be designed so operators can remain aware of automation bias tendency |
| SRC03-E01 | NIST AI 600-1 identifies confabulation and human-AI configuration risks including automation bias and overreliance |
| SRC01-E01 | CSET framework identifies user, technical, and organizational levels for addressing automation bias |
Contradicting Evidence¶
| Evidence | Summary |
|---|---|
| SRC04-E01 | Even the most recent FDA CDS guidance (January 2026) focuses on human independent review, not system output constraints |
Reasoning¶
The evidence consistently shows a pattern: regulated industries have identified the problem (automation bias, overreliance, inappropriate trust) but their solutions overwhelmingly focus on the human side (training, oversight, independent review). The closest system-side provisions are the EU AI Act's design requirements and NIST's risk framework, but these address transparency and interface design rather than directly constraining AI output generation. This represents an emerging but immature regulatory posture toward system-side requirements.
Relationship to Other Hypotheses¶
H2 occupies the middle ground between H1 (mature enforceable requirements exist) and H3 (no requirements exist at all). The evidence strongly supports this middle position — the problem is recognized, partial provisions exist, but no sector has produced specific output-behavior constraints.