Insights

Public Safety

When the voice on the line can't be trusted, verification has to live in the procedure

A new APCO and Public Safety Threat Alliance report, built on 766 survey responses from public safety communications professionals, documents call-takers running into AI-generated voices on swatting calls. The defense is written procedure and pre-arranged verification, not a better ear.

November 18, 2026 · 5 min read

When the voice on the line can't be trusted, verification has to live in the procedure

Swatting calls now arrive with synthetic voices, and the person answering cannot reliably hear the difference. APCO International and the Public Safety Threat Alliance's swatting report is based on 766 responses to their 2026 ECC Swatting Survey, and the full report says the survey "documented 262 instances where PSTs explicitly ran into AI-generated voice tactics, noting a signature robotic tone, flat cadence or unusual phrasing."

That robotic tone is the part that is going away. NPR quotes Mel Maier, APCO's CEO and executive director, saying swatting callers' voices "used to be pretty easy to recognize, but now they're natural sounding." If the cue a telecommunicator relies on is how the voice sounds, that cue has a short shelf life.

I am writing this as an advisor who works on AI risk in regulated settings, not as someone who has worked a dispatch console. The report and the people NPR interviewed are the authority here. My point is about where the defense has to live.

What the APCO and PSTA report found

The survey ran from December 4, 2025, to January 30, 2026, and drew 766 unique responses from APCO members and industry contacts. The report is careful about its limits: responses are self-reported, the survey is a point-in-time snapshot, and agencies define and log swatting differently.

Within those limits, the numbers describe a routine problem. Of the 766 respondents, 641 said they had to manage at least one swatting incident in the prior year. In the survey, 549 respondents flagged caller ID spoofing as the top barrier to verification. The report describes synthetic audio and pre-recorded background sounds, such as gunfire or screaming, as a growing part of the pattern, and says the background audio is the more dangerous of the two because it is built to push a telecommunicator past skepticism.

The report's sharpest finding for managers is not about technology. It says a significant number of ECCs have no written, swatting-specific SOPs for call handling, verification steps, supervisor notification, or coordination with responding agencies, so responses depend on individual experience or shift-level practice.

The constraint that shapes everything

Both people NPR interviewed say the same thing: dispatchers will treat any call reporting a life-threatening situation as real and send help. April Heinze of NENA put it plainly: "if it's a life-or-death situation, you cannot ever take chances on that."

That changes what verification is for. It is not a gate that decides whether to respond. It runs in parallel with the response and shapes how units arrive. The APCO report makes the same point in its own words: the assessments a telecommunicator makes in the first seconds of a call help decide whether field units "roll up with weapons drawn or approach with calculated caution."

So the question for a center is not "how do we detect AI voices?" It is "what can we confirm, quickly, that does not depend on trusting the voice?"

Verification that does not depend on the voice

APCO and NENA have chosen not to publish the specifics of their training, and NPR reports their reasoning: the more detail is public, the easier it is to train AI to get around it. I am going to respect that and stay at the level the report itself recommends publicly.

The report's recommendations fall into a few categories a center can act on:

  • Pre-arranged verification loops. Set up trusted contacts in advance with schools, campuses, large employers, and other likely targets, so a telecommunicator or supervisor can check the situation at the location while units are on the way. The report names school administrators and corporate security managers as examples.
  • Granular questioning. The report recommends questioning protocols that press for detailed physical facts about the location and the people involved, the kind a remote caller working from public information would not have. Maier told NPR that APCO is building training around questions at "such granular level that an AI cannot scoop that information from a public database."
  • Call-path awareness. Heinze told NPR that most swatters do not dial 911 directly, because 911 delivers location and callback data. Many calls reach police departments or other lines and are then transferred in. A procedure that notes how a call arrived gives the supervisor context, without changing the obligation to respond.
  • Supervisor escalation and field notification. A written step for when a supervisor is pulled in, and how concerns are passed to responding units so they can adjust their approach.
  • Logging. A consistent way to record suspected swatting calls, so patterns across shifts and jurisdictions can be seen at all.

None of these require detecting synthetic audio. They work the same whether the voice is human, recorded, or generated.

Write it down, then rehearse it

The gap the report keeps returning to is the absence of written procedure. That is fixable, and it is the part I would start with.

A useful SOP for suspected swatting is short. It says what triggers it, what the call-taker does in parallel with dispatch, when and how a supervisor is brought in, who the pre-arranged verification contacts are for the highest-risk locations in the jurisdiction, what is communicated to field units, and how the call is logged afterward.

Then rehearse it. The report recommends scenario-based training for telecommunicators and supervisors. A tabletop with dispatch, patrol, a school district contact, and the center's IT or cybersecurity lead can walk through a realistic call and find the steps that exist only in someone's memory.

Where to start

If your center does not have a written swatting SOP, draft one this month using the report's recommendations as the outline. If it does, check two things: whether the verification contact list for schools and major employers is current, and whether suspected swatting calls are being logged the same way on every shift. The voices will keep getting better. The procedure is the part you control.