What it is.
A flat list of alerts treats a person forcing a fence the same as a delivery driver at the front desk. Guards end up working the queue top to bottom and hoping the important one is near the top. Semaphor sorts the queue for them. Every card in the guard inbox carries one of four priority levels, and unresolved cards are always shown in priority order, oldest first within each level. It is how Semaphor turns a wall of cameras into clearer priorities.
| Level | How it gets there | What the guard does |
|---|---|---|
| Critical | The AI scene report flags clearly visible immediate danger, or the guard sets it. | Act now: pull the camera up, follow post orders, dispatch or call. |
| High | The AI flags activity that merits attention in this site’s context, or the guard sets it. | Review promptly and decide on a response. |
| Unassessed | A new preliminary alert that the AI has not yet reviewed, or has reviewed without escalating. | Look at the evidence; nothing has been ruled out. |
| Low | Set by the guard after review. | Routine activity, noted and deprioritized. |
Why it matters.
Work the right card first.
When five cameras light up during a busy stretch, the guard should not have to open all five to find the one that matters. Priority badges put it first.
Nothing gets quietly downgraded.
New alerts start as Unassessed, and an AI response of “no escalation” does not by itself push a card down or resolve it. Only a guard marks something Low. That conservative design protects against the worst outcome in monitoring: a real event quietly buried by software.
Supervisors can see the reasoning.
Every priority badge shows whether it came from the AI or from a guard, and why. Overrides carry an optional reason and survive restarts. At shift handoff or in a client review, the story of each card is on the card.
How it works.
Arrive as Unassessed.
An alert-first card lands immediately with its evidence, marked Unassessed.
AI suggests a level.
The AI scene report recommends urgent, review or no escalation, mapped to Critical, High or leaving the card Unassessed.
The guard decides.
The guard can set any level directly, with a reason. A guard decision cannot be overridden by an AI report about older frames.
Escalation rules.
- Higher priority reopens attention. If a card the guard acknowledged is raised by genuinely newer evidence, it comes back to the top of the queue.
- AI never lowers. The AI can raise priority; it never lowers a card or closes it.
- Priority orders AI review too. Pending scene reports are processed in priority order, so a Critical card gets its report ahead of a Low one.
- Every camera still gets watched. Priority orders the review queue, not observation. Camera sampling stays fair across every view.
Automated severity scoring, in development.
Today’s levels are triage suggestions shaped by your site policy and your guards’ overrides. Our engineers are building the next layer: automated severity scoring evaluated against site-specific rubrics, independently labelled events and held-out cameras, so a Critical badge means exactly what your post orders say it means. Pilot partners get it first — see what we’re building next.
Where it matters.
- Campuses and offices — hundreds of cameras, mostly routine daytime traffic, and a handful of events that need someone now.
- Logistics hubs — constant vehicle movement where the priority is the truck at the wrong gate after hours.
- Data centers — strict perimeters where a person at a fence line should outrank everything else on the wall.
For how to design levels that match your post orders, read security alert triage and priority levels.
Frequently asked questions.
Who decides an alert’s priority, the AI or the guard?
Both contribute, but the guard wins. The AI suggests a level from its scene report; the guard can set any level directly, and an AI result about older frames cannot override that decision.
Why is there an Unassessed level?
Because a new alert has not been ruled out. Unassessed keeps fresh activity visible until either the AI or a guard has actually looked at it.
Can priorities be tuned to our site?
Yes. A site policy drawn from your post orders shapes the AI’s recommendations, and we calibrate levels to your operation during a pilot.
See your own alerts ranked: book a live demo or start a pilot.