Hii Research Audit v0.2
The source-level record supporting the provisional v0.2 revision. Each entry identifies the source, its evidentiary role, and the limitations carried into the audit.
Verified Sources
Author or organization: European Commission
Publication date: 2026, last updated
Source type: Official government website
Claim supported: Recorded as the source for the staged implementation timeline of the EU AI Act.
Quantitative details: Not applicable.
Limitations: The underlying page did not provide a clear original publication date and may continue to change.
Author or organization: UNESCO
Publication date: 16 May 2023, updated 26 September 2024
Source type: Intergovernmental organization article
Claim supported: Confirms that UNESCO’s Recommendation on the Ethics of AI was adopted in November 2021 and applies across UNESCO member states.
Quantitative details: Not applicable.
Limitations: The page did not substantiate the readiness-assessment count used in the earlier draft.
Author or organization: Hailey Onweller, Elias Lumer, Austin Huber, Pia Ramchandani, Vamse Kumar Subbiah, and Corey Feld
Publication date: 7 May 2026, arXiv preprint
Source type: Computer-science preprint
Claim supported: Used to examine the difference between valid links, topical relevance, and factual accuracy in deep-research systems.
Quantitative details: The paper evaluated 14 closed- and open-source models using rubric-based measures and reported declining fact-check accuracy as retrieval depth increased.
Limitations: Preprint status, reliance on automated judgments, and no independent human ground-truth dataset.
Author or organization: Hannah Rose Kirk, Henry Davidson, Ed Saunders, Lennart Luettgau, Bertie Vidgen, Scott A. Hale, and Christopher Summerfield
Publication date: 1 December 2025, revised 18 February 2026
Source type: Human-computer interaction preprint
Claim supported: Used for the report’s discussion of a four-week randomized study of relationship-seeking AI and the distinction between liking, wanting, attachment, and psychosocial benefit.
Quantitative details: Reported sample size of 3,534 participants over four weeks.
Limitations: Short duration, preprint status, and limited demographic detail in the audit record.
Author or organization: AI & Psychotherapy
Publication date: Accessed July 2026
Source type: Independent mental-health technology review
Claim supported: Used for descriptive claims about Woebot’s design, trial history, user reach, regulatory designation, and reported reasons for retirement.
Quantitative details: The secondary source described multiple small trials and a large observational dataset.
Limitations: Secondary reporting, not peer-reviewed, and some totals may be estimates rather than official counts.
Author or organization: Woebot Health
Publication date: 2025
Source type: Company FAQ
Claim supported: Confirms the retirement date of the Woebot app and that new accounts could no longer be created.
Quantitative details: Not applicable.
Limitations: Confirms retirement, but does not independently establish the reasons for shutdown.
Author or organization: National Institute of Standards and Technology
Publication date: 26 January 2023, with later updates
Source type: Official U.S. government website
Claim supported: Used for release dates and the report’s description of the voluntary AI Risk Management Framework and later generative-AI guidance.
Quantitative details: Not applicable.
Limitations: The page provides framework information and dates, not comprehensive adoption data.
Author or organization: Anokhy Desai, International Association of Privacy Professionals
Publication date: 23 August 2023
Source type: Professional association article
Claim supported: Used to describe model cards as documentation tools and to qualify the maturity and non-mandatory status of model-card practices.
Quantitative details: Not applicable.
Limitations: Industry commentary rather than peer-reviewed research.
Author or organization: Coalition for Content Provenance and Authenticity
Publication date: 2024
Source type: Industry consortium website
Claim supported: Used to describe C2PA as an open standard for cryptographic content credentials and provenance metadata.
Quantitative details: Not applicable.
Limitations: Promotional overview without empirical adoption or effectiveness data.
Author or organization: Hybrid Intelligence Institute
Publication date: 2026
Source type: Internal methodological guide
Claim supported: Provided the governing methodology for evidence labels, skeptical design, mechanism transparency, longitudinal measures, and correction.
Quantitative details: Not applicable.
Limitations: Internal guidance and not external evidence.