Self-Harm and AI

Incidents linking AI and self-harm, how AI may have contributed, and what oversight policies model providers and third parties had in place at the time.

Explore the archive

• Full Paper
• About the Archive
• Abstract
• 1 Introduction
• Part I: Incident Registry and Methods
  • 2 Incident Registry
  • 3 Failure-Mode Taxonomy
  • 4 Methods
  • 5 Evidence Grades
  • 6 Coding Rules
  • 7 Source Selection
• Part II: Evidence and Governance
  • 8 Incident Directory
  • 9 Provider Index
  • 10 Policy Documents
  • 11 Appendices
  • 12 Data and Reproducibility Files
• PDF Edition
• License

Scope and limits

The archive separates public evidence, coded records, derived analysis, and explanatory mechanisms. It does not establish prevalence, rank platforms by risk, or treat a coded association as proof of legal or clinical causation.

Cite this work

To cite this archive, use the following BibTeX entry:

@dataset{roberts2026selfharmai,
    title     = {Self-Harm and AI: An Incident Registry, Taxonomy, and Governance Framework},
    author    = {Roberts, Brian J.},
    publisher = {OSF},
    year      = {2026},
    version   = {2.0},
    doi       = {10.17605/OSF.IO/49AGJ},
    url       = {https://selfharm.ai}
}

A machine-readable citation is available as CITATION.cff. The archival DOI is 10.17605/OSF.IO/49AGJ.

Release history

• [v1.0] Thu, 8 Jan 2026 10:56:14 PM UTC
• [v2.0] Wed, 25 Mar 2026 2:02 PM UTC — current public version