VARNIUMAI risk defense / everyday bearings

DEEPFAKE / PUBLIC ALERTS

A Message Says Evacuate Now. Check Which Channel It Arrived On.

Synthetic audio, official-looking graphics, and urgent wording are all cheap. The channel a public warning arrives on is not — real alerts come through a defined system, and the fastest check is whether the same warning is also on that system.

Start here

The free tool below runs entirely in this browser — no account, nothing stored or transmitted. Redact anything you would not want kept before you type it.

Open the companion tool →

Why the channel beats the audio

Public warning systems are defined by delivery, not by production. In the United States, Wireless Emergency Alerts reach your phone through the national alerting system with a distinctive tone and vibration cadence, are geographically targeted by an authorized government alerting authority, and do not ask you to pay or log in. Broadcast stations carry Emergency Alert System messages, and NOAA Weather Radio carries the same warnings on a dedicated receiver. A social post, a forwarded screenshot, a robocall, an email, or a voice message from a number you do not know is not one of those channels, however it sounds.

That distinction survives better synthesis than any content check. A generated voice can imitate an official tone and a generated graphic can imitate an agency logo, but neither can make a message appear on the alert system, on the agency's own page, and on the weather radio simultaneously. Confirming against two official sources takes a minute and requires nobody to judge the audio — which is the part that is getting harder.

The protocol

  1. Note how the message reached you: a Wireless Emergency Alert on your phone, a TV or radio broadcast, the agency's own page or account, or something else — a social post, a forwarded screenshot, a robocall, an email, or a voice message from an unknown number.
  2. If you are in immediate danger, or the message tells you to leave or take cover now, act on the protective instruction first and verify while you move. Delaying evacuation to fact-check is the wrong order.
  3. Confirm against a second official channel: the responsible agency's own published page or account, a local broadcast station, or NOAA Weather Radio.
  4. Look at what the message asks you to do. Official alerts do not ask for payment, do not ask you to log in or register, and do not send you to a shortened link to claim assistance.
  5. Check the source rather than the share: who posted it, whether the account is the agency's own, and whether the same warning appears on the agency's website.
  6. If it is not confirmed on an official channel, do not reshare it. Report the account or number to the platform, and report it to the FTC.

What the tool does

Describe how an urgent alert reached you and what it asked you to do, and get a channel reading: which delivery paths are official, which are not, and the two sources to confirm against. It runs in this browser; nothing is sent or stored.

Limit first: This is a channel-check aid, not an authenticity verdict, and it cannot prove a message is real or fake. It is never a reason to delay protective action — if you are told to evacuate or take cover, do that first and verify while you move.

Open the companion tool →

One human next step

For an immediate threat to life, call 911 (US) — do not wait for a check to resolve. For everything else, contact the agency named in the message through the number or page it publishes itself: local emergency management offices answer questions about alerts for their area, and an unconfirmed alert can be reported to the platform that carried it and to the FTC at ReportFraud.ftc.gov.

Research log and safety checks

AI-assist path: not used for a verdict, and deliberately so. No model can separate a synthetic broadcast from a real one reliably enough to act on, and a false 'real' verdict here could delay evacuation — the failure mode §0.5d and §1.2 UIC-10 exist to prevent. The reading is a channel inspection plus two official confirmations, which is stronger than any audio judgment. §6.9 disclosed: audio-deepfake detection is an active field but is not achievable in-browser at zero cost with an honest basis, so the tool routes to channel provenance instead. §6.10: not a frontier-pace topic.

§0.5 protection result: category is deepfake-media-verification; crisis-adjacent: no; fear/urgency/scarcity toolkit used: no. The protective function is free and the paid feature is convenience only.

Related dispatches

Pairs with the same protection bar: a free complete reading, an optional convenience unlock, and no fear-framing.