Internal policy · Verified
Archive status disclaimer
Added a clear provisional-draft disclaimer and made evidence status part of the report’s visible publication boundary.
Hii Research Audit v0.2
Revision Log
A public record of the main corrections, qualifications, methodological additions, and structural changes made during the v0.2 revision.
Status: Provisional supporting document · July 2026
Change Record
What changed and why.
Recorded
European Commission AI Act page · Verified
EU AI Act timeline
Corrected the timeline to distinguish entry into force, staged obligations, full applicability, and later high-risk-system dates.
UNESCO article · Verified; readiness count unavailable
UNESCO Recommendation
Confirmed the 2021 adoption of the Recommendation and removed an unsupported readiness-assessment count from the main narrative.
Kirk et al. preprint · Verified as source
Longitudinal randomized study
Added a qualified summary of the four-week relationship-seeking AI study and retained its preprint limitations.
Company FAQ · Verified; secondary review · Emerging Evidence
Woebot details
Separated confirmed retirement information from secondary claims about trial totals, user counts, regulatory designation, and shutdown causes.
Hii Epistemic Guardrails · Internal standard
Hii Research Reliability Standard
Added explicit evidence labels and bibliographic and methodological requirements for claims and citations.
Hii Epistemic Guardrails · Internal standard
Relational AI Evidence Standard
Defined relational evidence and the methodological requirements of longitudinal human-AI relationship research.
Gap analysis and Hii methods · Internal standard
Memory and Continuity Disclosure
Added disclosure requirements covering memory activation, memory sources, deletion options, model versions, and project context.
Hii Epistemic Guardrails · Internal standard
Methodological standards
Added pre-registration, independent observation, mechanism transparency, skeptical design, relational measures, and continuity-as-reconstruction.
Gap analysis · Verified as internal revision
Failure modes
Expanded failure modes to include memory misalignment, retrieval failure, project-boundary confusion, user mental-model gaps, model-change opacity, citation fallibility, and interface instability.
Hii standards · Internal revision
Recommendations
Integrated research reliability, provenance, AI research disclosure, verification workflow, confidence rating, relational evidence, and memory-continuity disclosure.
Hii internal case · Verified
Case Study Zero
Preserved an internal audit failure involving a fabricated statistic as a permanent methods case and a reason to strengthen verification.
Primary and professional sources · Verified or Emerging Evidence
Citation and provenance claims
Re-verified or reclassified claims about citation systems, retrieval-augmented generation, C2PA, model cards, and the NIST AI RMF. Unsupported claims were softened, flagged, or removed.
Editorial revision · Verified
Table of contents and cross-references
Updated the report structure to include the new methodological section and reorganized insertions.