AI Governance and Public Safety Technology in Hawaii

Summary of the paper

This Hawaii Policy Research paper asks which AI governance gaps Hawaii still faces after the 2026 Legislature's AI measures, specifically for public safety AI tools and how well they are calibrated for Native Hawaiian and Pacific Islander communities. It reviews the 2026 Hawaii Legislature's AI governance measures, the NIST Face Recognition Vendor Test findings on demographic effects, behavioral risk assessment validation research and ISPI's AI governance and behavioral threat assessment frameworks as applied to Hawaii's demographics and jurisdictions.

The paper's core concern is that public safety AI tools trained on continental data may be systematically miscalibrated for Native Hawaiian, Pacific Islander, Micronesian and Polynesian populations, and that the harm can go unnoticed because these populations are too small for standard audit methods to detect. Its answer is to move from after-the-fact detection to validation before deployment.

What the paper finds

• The 2026 measures address elections, not public safety. The Legislature targeted AI-generated political content meant to deceive voters, which the paper calls a meaningful first step, but it left the calibration of public safety AI for Native Hawaiian and Pacific Islander communities unaddressed.
• Small populations create an audit gap. Standard algorithmic audits find bias through statistically significant differences in large datasets. Case volumes for Native Hawaiian and Pacific Islander people in a single deployment may never reach those thresholds, so a miscalibrated tool could run for years without triggering detection.
• Risk assessment instruments need local validation. Instruments used in Hawaii Family Court, parole and pretrial detention were developed and validated mainly on continental populations. The paper says scores produced without Native Hawaiian and Pacific Islander validation should be treated as provisional.
• Predictive policing could amplify existing disparities. Systems trained on Hawaii's historical arrest data may encode the decision patterns ISPI links to disproportionate Family Court involvement for Native Hawaiian and Pacific Islander youth.
• No rule requires pre-deployment validation. No current regulatory framework requires tools to be validated for these populations before they are used.

What it recommends

• Hawaii State Legislature: require pre-deployment demographic validation for all AI public safety tools deployed by Hawaii government agencies, including vendor disclosure of accuracy and calibration metrics for Native Hawaiian and Pacific Islander populations before deployment is authorized.
• Hawaii law enforcement agencies: make documented demographic validation a mandatory part of any AI tool procurement, with contract terms for ongoing accuracy reporting and recalibration.
• Hawaii Judiciary: commission an independent audit of whether risk assessment instruments used in Family Court, parole and pretrial detention produce calibrated scores for these populations.
• ISPI and the University of Hawaii: commission research on Pacific Island behavioral norms to support calibration of both behavioral threat assessment (BTAM) and AI tools.

Limitations the paper notes

Systematic empirical research on AI public safety tool miscalibration in Native Hawaiian and Pacific Islander contexts is limited. The paper's characterization of the governance gap is derived from ISPI's analytical framework and NIST research.

Who it's for

Hawaii legislators and committee staff working on AI policy, police and public safety procurement officers evaluating AI vendors, judiciary and corrections officials who rely on risk assessment scores, and community advocates concerned with Native Hawaiian and Pacific Islander justice outcomes.

Cite as

Pulley, Warren (2026). AI Governance and Public Safety Technology Policy for Hawaii: Native Hawaiian and Pacific Islander Population Validation, Small-Population Bias Detection, and Island-Specific Accountability (ISPI HI-10). Island Security Policy Institute, Honolulu. https://ispiglobal.com/papers-landing/hi10-ai-governance-hawaii https://doi.org/10.5281/zenodo.23449961

@techreport{ispihi102026,
  author      = {Pulley, Warren},
  title       = {AI Governance and Public Safety Technology Policy for Hawaii: Native Hawaiian and Pacific Islander Population Validation, Small-Population Bias Detection, and Island-Specific Accountability},
  institution = {Island Security Policy Institute},
  address     = {Honolulu, HI},
  number      = {ISPI HI-10},
  year        = {2026},
  url         = {https://ispiglobal.com/papers-landing/hi10-ai-governance-hawaii},
  doi         = {10.5281/zenodo.23449961}
}

Related analysis