Yellow Light — What Needs Your Review
Yellow light tools have real value — but they require a mandatory human review step before their output can be acted on, signed, or submitted.
Most of Your AI Tools Live Here
Yellow light is the most important category in the Traffic Light Framework — because it’s where most of the AI tools in public safety actually belong.
Yellow doesn’t mean dangerous. It means: valuable with guardrails.
The mandatory review step isn’t bureaucratic friction. It’s the mechanism that makes these tools legally defensible. Without human review built into the workflow, a yellow-light tool becomes a red-light situation.
Yellow Light Use Cases
BWC Summarization — For Training Only
Body-worn camera footage can be processed by AI to generate summaries for after-action review, training purposes, and quality assurance. This is yellow — not red — when the summary is clearly labeled as an AI draft, reviewed by a supervisor, and never used as a substitute for the actual footage in an investigation.
The mandatory review: A supervisor or trainer must review the AI summary against the actual footage before it’s used in any training context or performance review.
Why not green: BWC footage is CJIS-protected evidence. AI processing of this footage requires specific compliance controls. The summary could contain errors that affect how an officer’s conduct is evaluated.
AI-Suggested (Not AI-Decided) EMS Triage
AI tools like Corti that suggest triage priority — “this call has indicators consistent with cardiac arrest” — are yellow when the dispatcher uses the suggestion as input to their own clinical judgment, not as a final decision.
The mandatory review: The human dispatcher evaluates the AI’s flag against everything else they’re hearing and makes the triage decision.
Why not green: The stakes are life-and-death. A missed cardiac arrest or a misclassified medical emergency can be fatal. The AI’s role is to assist the dispatcher, never to replace their judgment.
Preliminary Thermal Drone Analysis
AI-assisted analysis of thermal drone footage for fire mapping, search and rescue, or scene assessment is yellow. The AI can rapidly identify heat signatures and flag areas of interest — but the interpretation of what those signatures mean requires human assessment.
The mandatory review: An incident commander or qualified operator reviews the AI analysis before tactical decisions are made based on it.
Why not green: Tactical decisions based on incorrect heat signature interpretation can put crews in danger.
Draft Reports — With Mandatory Redline
This is the big one. AI-drafted incident reports are yellow — they’re valuable, they save time, and they’re going to become standard. But every draft requires a documented line-by-line review before an officer signs it.
The mandatory review: Officer cross-checks the draft against BWC footage and personal recollection. Any unverifiable claim is removed or corrected. Officer signs only what they can personally attest to.
Why not green: The output is a legal document with criminal case implications. An error that slips through without correction becomes the officer’s legal attestation.
What “Mandatory Redline” Actually Looks Like
A redline review isn’t skimming for obvious errors. It’s a structured process:
- Read the draft aloud — your brain catches things your eyes miss when reading silently
- Flag every direct quote — verify each one against BWC audio
- Flag every specific claim — addresses, times, vehicle descriptions, physical descriptions
- Flag every action attributed to you — anything you allegedly said, did, or observed
- Check the disposition — make sure the final status, charges, and evidence notes are accurate
- Cross-reference with CAD — confirm times and incident numbers match
This process takes 5–15 minutes depending on incident complexity. It replaces 60–90 minutes of writing from scratch. The math still works overwhelmingly in your favor — as long as you actually do it.
Every department will eventually have its own AI report review protocol. Before that’s handed to you, build your own working version.
Using the five-step process above as a starting point, create a checklist specific to the types of calls you handle most. For each item, note:
- What you’re checking for
- Where you verify it (BWC footage, CAD log, personal notes)
- What you do if it doesn’t match
Example for a domestic disturbance call:
| Check | What I’m Looking For | Where I Verify |
|---|---|---|
| Party names and DOBs | Correct spelling, correct person | CAD lookup, ID documentation |
| Direct quotes | Exactly what was said | BWC audio, timestamp |
| Physical actions | Who did what, in what order | BWC video |
| Weapons mentioned | Any reference to weapon present or absent | BWC + my own notes |
| Disposition | Charges, transport, notifications | Supervisor confirmation |
Build your version for your most common call types. This is your personal AI review protocol.
The most dangerous moment for a yellow-light tool is a high-volume shift.
When you’re on your seventh call of the night, the AI draft looks fine, you’re already mentally on the next scene, and the review step feels like friction — that’s exactly when you’re most likely to sign something you shouldn’t.
Yellow means yellow regardless of how your shift is going. The review step is not optional when you’re tired. If anything, fatigue is when you need it most — because your memory and attention are at their lowest exactly when the AI is most likely to slip something past you.
An officer's department has just implemented an AI report drafting tool. There is no formal policy yet on review requirements. The officer drafts reports with AI and does a quick read-through before signing. Is this sufficient?
Reveal Answer & Explanation
No — a quick read-through is not a redline review.
Reading for general coherence misses most AI hallucinations. The human brain fills in expected information when reading quickly — especially when tired and when the general narrative matches what you remember from the scene.
Regardless of whether a formal policy exists, the legal standard doesn’t change: an officer who signs an incident report is attesting to its accuracy. That standard requires more than a quick scan.
Until a formal policy is in place, officers should apply the most rigorous review standard available — treating every AI draft as a yellow-light document requiring structured redline review. This protects the officer legally and demonstrates the professional standard the department should eventually codify.
Up Next
Yellow light tools are where the work happens — and where training makes the difference. Red light is where the line is drawn completely.
Lesson 3: Red Light — what’s strictly off-limits, no exceptions.