AI governance in public safety cannot live only in a policy binder. It has to guide how tools are identified, reviewed, approved, used, supervised, evaluated, explained, and retired.
Start with inventory. Leaders need to know which systems use AI-enabled functions, where those functions appear in the organization, what data they touch, who relies on their output, and whether personnel are using unapproved tools outside formal systems. Without that baseline, oversight begins after adoption instead of before it.
Define the purpose and boundary of each use. What task does the tool support? What may it recommend, summarize, generate, detect, or prioritize? What is it not allowed to decide? Who remains accountable for the action? What evidence would show that the use is valuable, and what conditions would require it to pause?
Match review to risk. A low-impact administrative aid should not require the same process as a system that may influence enforcement, identification, evidence, deployment, or access to services. Higher-impact uses may require deeper legal, operational, technical, records, security, privacy, civil-rights, labor, procurement, and community review.
Human oversight must be specific. "A human is in the loop" is not enough. The organization should define who reviews an output, what they are expected to check, what information they need, how disagreement is recorded, and who owns the final decision. Personnel also need a clear way to report questionable output or unintended effects.
Training and monitoring turn policy into practice. Users should understand a tool's intended purpose, limitations, prohibited uses, documentation duties, and escalation path. After launch, leaders should review performance, errors, user behavior, vendor changes, data practices, complaints, and whether the tool continues to serve the approved purpose.
Every approval should include an exit path. A tool may need to be modified, restricted, suspended, or retired when it fails to deliver value, creates unacceptable risk, changes materially, or no longer fits the mission. Responsible adoption includes the ability to stop.
Governance is not a promise that every risk will disappear. It is a disciplined way to make decisions visible, assign accountability, and respond when evidence changes.
By Darrell Lowe, Founder, The Lowe Group.AI, LLC