AI doesn’t just automate decisions—it creates new exposure.

Public safety leaders are increasingly evaluating AI for dispatch, analytics, transcription, and investigative triage. But few have built the governance frameworks needed to manage the risks that come with these tools. Here’s what every agency should be watching for.

‍

1. No Policy

Without a written and enforceable AI policy, your agency lacks a framework for oversight, training, or escalation. That’s the foundation—and too many agencies skip it.

2. No Risk Classification

Treating all AI tools the same leads to overregulation of low-risk tools and dangerous underregulation of high-risk ones. You need a risk-tiering framework.

3. No Human-in-the-Loop Review

Systems making or influencing decisions need to be reviewed—and logged—by trained personnel. “We trusted the system” isn’t defensible.

4. No Logs

If your agency can’t produce audit trails of system access, outputs, and review decisions, you can’t defend your use—or improve it.

5. No Vendor Oversight

Letting vendors define what counts as AI—or what data they retain—creates procurement and legal risk. Your contract must lead the relationship.

6. No Incident Plan

What happens if something fails? If the public files a complaint? Agencies without kill switches, review logs, or disclosure plans get caught flat-footed.

7. No Transparency

If the community, council, or oversight board doesn’t know what tools you’re using or how they’re governed, trust erodes. And media stories fill the vacuum.

8. No Internal Training

You can’t deploy AI tools if staff don’t know how to use them—or how to flag errors. Training should match your policy and risk level.

9. No Legal Alignment

Is your AI use compliant with local, state, and federal standards? Can you answer a FOIA request with confidence? If not, pause.

10. No Certification or Audit Readiness

You don’t have to be perfect—but you need to show that your agency has a process in place, is documenting its decisions, and is ready to respond to questions.