NIST has documented demographic differentials in face recognition error rates. In ISPI's analysis, small island jurisdictions need validation, auditing and disclosure suited to small populations.
Abstract
Examines the deployment of AI public safety technology in island and Pacific Islander communities, examining calibration risks and proposing an accountability framework for equitable AI deployment in SIDS contexts. All ISPI research is free under Creative Commons Attribution 4.0.
Key Findings
NIST found demographic differentials in error rates for the majority of face recognition algorithms tested (NISTIR 8280, 2019)
In ISPI's analysis, low case volumes make bias harder to detect in small jurisdictions
ISPI proposes pre-deployment validation, independent audits, vendor disclosure and individual review, building on the NIST AI Risk Management Framework
Summary of the paper
"AI Governance for Public Safety Technology in Island Communities" sets out a governance framework for AI public safety tools in small island and Pacific Island jurisdictions.
It builds on NIST's 2019 face recognition demographic effects study (NISTIR 8280) and the NIST AI Risk Management Framework (2023).
What the paper finds
- Differentials are documented. NIST found demographic differentials in error rates for the majority of face recognition algorithms tested.
- Small population accountability gap, in ISPI's analysis. Low case volumes make bias harder to detect, and oversight is thinner.
- Behavioral scoring needs cultural calibration, in ISPI's analysis, so culturally normal communication is not read as a risk indicator.
What it recommends
- Governments: pre-deployment validation against local data, with individual review and appeal.
- Federal agencies: support independent audits with Pacific Island expertise.
- Vendors: disclose demographic accuracy data.
- ISPI and academic partners: Pacific Island behavioral baseline research.
Who it's for
Public safety agencies, technology procurement officers, oversight bodies, vendors, and researchers.
Pacific Island governments implementing AI public safety technology may face similar calibration risks, in ISPI's analysis. Discipline 14 — Decolonizing Research — is mandatory for all research in this domain.
Cite as
Pulley, Warren (2026). AI Governance and Public Safety Technology Policy for Island and Small-State Communities: Accountability, Bias, and the Missing Framework for Small Jurisdiction AI Deployment (ISPI WP-11). Island Security Policy Institute, Honolulu. https://ispiglobal.com/papers-landing/wp11-ai-governance https://doi.org/10.5281/zenodo.23449816
BibTeX
@techreport{ispiwp112026,
author = {Pulley, Warren},
title = {AI Governance and Public Safety Technology Policy for Island and Small-State Communities: Accountability, Bias, and the Missing Framework for Small Jurisdiction AI Deployment},
institution = {Island Security Policy Institute},
address = {Honolulu, HI},
number = {ISPI WP-11},
year = {2026},
url = {https://ispiglobal.com/papers-landing/wp11-ai-governance},
doi = {10.5281/zenodo.23449816}
}Related Research
HI-10 — AI Governance Hawaii →WP-10 — Democratic Resilience →CM-04 — AI Public Safety Island Communities →Related analysis
AnalysisHPD Chief Lazar Announces Drone Program. The AI Governance Question He Hasn't Answered Yet. →AnalysisBehavioral Threat Assessment in Island Communities — Why Cultural Calibration Changes Everything →AnalysisNative Hawaiian and Pacific Islander Youth in the Justice System — Decades of Disparity, Decades Without Research →PaperWP-12 — Corporate Security and ESG Risk Policy for Island Operations →GuideHawaii Law Enforcement Workforce & Recruitment: Research and FAQ →Pulley, Warren. "AI Governance and Public Safety Technology in Island Communities." Island Security Policy Institute, 2026. https://ispiglobal.com/papers-landing/wp11-ai-governance
