A CONVINCING FAKE EMERGENCY WARNING CAN TRIGGER REAL DANGER. LEARN HOW TO VERIFY BEFORE YOU ACT.
The next emergency alert you see might be fake—and it may look completely convincing. AI-generated photographs, cloned voices, synthetic video and altered screenshots can spread during power cuts, extreme weather, cyber incidents or international crises. Even genuine footage can mislead when it is old, filmed somewhere else or paired with a false caption. Acting on the wrong information can be more dangerous than the original event.
In April 2025, independent fact-checking organisation Full Fact published a practical guide explaining why it is difficult to identify artificial images, video and audio from visual clues or automated detectors alone. The Bellingcat Online Investigation Toolkit provides open-source resources for checking where material originated, tracing earlier appearances and assessing locations. Both approaches emphasise evidence beyond an image's immediate appearance.
Here is a practical method for deciding what to trust when a convincing clip claims a dam has failed, a hospital is evacuating or a government has issued an urgent warning. The first step is surprisingly simple: do not let the emotional force of a video make the decision for you.
Why AI misinformation is especially dangerous during emergencies
In normal conditions, a false post may waste time or damage someone's reputation. During an emergency, a false instruction can send people toward a hazard, create unnecessary travel, overload emergency lines or lead families to overlook the real notice from their service provider.
AI changes how cheaply and quickly convincing material can be produced. A fabricated announcement may include a plausible voice, an invented newsreader, familiar branding and a false screenshot of an authority's social-media account. Large volumes of similar posts can make an invented claim appear independently confirmed.
But not all misinformation is AI-generated. An authentic video from a storm two years ago can be reposted as “LIVE NOW”. A real photograph can be paired with a fabricated location. A genuine official statement can be clipped until its meaning reverses. Those techniques remain effective precisely because people concentrate on whether the image looks fake instead of whether the claim is supported.
The four questions that matter more than spotting AI fingers
When you see a potentially urgent image, video or voice clip, begin with four checks:
| Question | What to look for | Possible warning sign |
|---|---|---|
| Who published it first? | Original uploader, agency, eyewitness, date | Only reposted accounts repeat the same clip |
| Where was it recorded? | Landmarks, road layout, official location | Background does not match the claimed place |
| When was it recorded? | Earliest upload, weather, daylight, event history | Same footage existed before the reported incident |
| Who independently confirms it? | Service provider, emergency authority, reputable reporting | Many posts all trace to one unverified account |
Full Fact's guide warns against treating an AI detector or a strange-looking detail as a final verdict. A video without obvious glitches can still be synthetic, and genuine recordings can contain compression artefacts or misleading captions.
A fast verification routine when a crisis claim appears
Use two passes: a short safety check if the claim could affect you immediately, followed by a more careful investigation if you have time. You do not need to become a professional investigator to make better decisions.
Step 1: Stop forwarding it
Before sharing, ask what the post is telling people to do. If it urges immediate travel, withdrawal of money, a fuel purchase or abandoning a safe location without confirming the source, treat it as unverified. Forwarding an unverified warning to a family group can make others act before you have checked it.
Step 2: Check the direct official channel
If a post claims the electricity supplier has ordered an evacuation, find the supplier's known official website or contact number directly. Do not follow a shortened link in the suspicious message. If it purports to be an emergency alert, compare it with your country's recognised alert service or the relevant local authority.
Important safety exception: if you can personally see smoke, floodwater or another immediate danger, do not delay protective action to complete an online fact-check. Follow emergency-service instructions and use local emergency numbers when appropriate.
Step 3: Find the earliest visible copy
Search for distinctive words from the caption and for the location or organisation named. A reverse-image search may show that the same photograph was published long before today's alleged event. For video, find a distinctive frame and compare it with earlier posts and credible news footage.
Step 4: Check place and time
Look for persistent features: building shapes, signs, roads, skylines and terrain. These can be compared with reliable maps or other photographs. Then check whether the daylight, season, weather and event chronology make sense. One mismatch may be explained by camera angle or editing; several independent mismatches are more revealing.
Step 5: Look for independent confirmation
Does a known local authority confirm the disruption? Are other witnesses recording it from different positions? Has a credible newsroom verified the scene? Ten accounts reposting the same source do not count as ten independent confirmations.
Step 6: State the uncertainty honestly
Sometimes you cannot establish whether the material is genuine. The safe conclusion is unverified, not automatically false or AI-generated. Record what you checked, avoid sharing unsupported instructions and follow confirmed information as it becomes available.
Case study: a fake water contamination warning
The following is a fictional training example, not a real incident.
A dramatic screenshot appears in a neighbourhood group: “Emergency: tap water is unsafe across the whole county. Fill your car and leave tonight.” The logo resembles a water company and the wording seems official. A voice note claims the announcement was heard on local radio.
Instead of treating the message as an order, the household checks the water company's official status page and the local authority's emergency updates. Neither carries the claimed instruction. Searching a distinctive sentence reveals only copies of the same screenshot. The person forwarding the voice note cannot identify the original broadcast. The claim remains unverified.
The family does not ignore the possibility of a genuine water problem. It keeps normal emergency drinking-water supplies, watches the direct provider channel and follows any real safety notices. That distinction is crucial: verification is not denial of danger. It is a way to respond to the right danger.
How can AI clone a trusted voice?
Modern tools can imitate vocal characteristics using recorded examples. A caller may sound like a relative, colleague, employer or public official. The danger comes when familiarity with the voice overrides the usual check that a request makes sense.
Suppose you receive a rushed message apparently from a family member demanding money because transport has stopped. Do not send money solely on the basis of the voice. Call the person back using a number you already know, or use a previously agreed separate channel. If the caller pressures you not to verify, treat that as an additional warning sign.
A simple household practice helps: agree in advance how family members will check an unexpected urgent request. That could be a return call to a known number plus a second contact if necessary. Avoid sharing private verification details publicly or treating one shared word as an infallible security system.
Can an AI detector tell you whether a video is fake?
Not with certainty. Some AI-detection tools may offer useful signals, but they can produce false positives on authentic material and miss manipulated images. Content may also be compressed, cropped, translated or resaved in ways that complicate detection.
Full Fact explains that automated AI-detection tools can be unreliable and that its investigators prefer established verification techniques: reverse-image searches, contextual clues and consultation with specialists. Bellingcat's toolkit also lists tools for image and video verification, geolocation and searching earlier copies of content. An apparently confident detection result should be weighed against these independent checks.
A label such as “98% likely AI” is not a substitute for verifying the underlying event. Equally, a video that passes an AI detector may be real footage used to tell a completely false story.
How misinformation could worsen an AI-related disruption
Imagine a real cyber incident at a regional service provider. Some customers cannot access accounts, but essential services remain operational. False posts claim the whole transport system has collapsed and tell everyone to leave home. Synthetic footage appears to show crowds and official announcements. People rush toward shops or roads based on invented details.
That is a plausible scenario for a preparedness exercise; it is not a report of an actual incident. The risk is amplified by uncertainty. When official websites are overloaded or communications are interrupted, rumours may occupy the information gap.
Our investigation of AI and critical infrastructure explains why real service failures can affect several linked systems. The safest information plan has both digital and non-digital routes: the service provider's official announcements, verified local reporting and a battery or wind-up radio where available.
Build your household's trusted-information plan
Do this before an emergency rather than during a flood of alarming messages.
- Pick your essential local sources. Save the correct electricity network operator, water supplier, local emergency management authority and recognised weather-warning service.
- Store direct links and phone numbers. Keep them on paper as well as on your phone so a social-media post cannot become the only route to an update.
- Agree a family rule. Nobody forwards a major emergency instruction without saying who issued it, where it was confirmed and when it was published.
- Choose a fallback. If the internet is unavailable, know which radio station or authorised public information source serves your area.
- Rehearse with one example. Find a viral claim from a past event and see whether you can identify its original date and location without relying on an AI detector.
Keep the rules realistic. Do not require an elderly relative to become a digital investigator. A trustworthy direct contact, clear paper instructions and one person designated to verify messages may be more useful.
For the practical service-disruption response these checks support, see How to Prepare for an AI-Related Infrastructure Crisis. During a conflict, our autonomous weapons guide also explains why viral claims about unfamiliar military technology should be verified before being repeated.
What to do when sources disagree
Two apparently reputable sources can report different facts in a fast-changing event. One may be correct but old; another may cover a smaller area; a provider may confirm the existence of an incident without knowing its cause.
Separate the statements: what happened, where, when, what is confirmed, and what action is being recommended? Prioritise the most recent relevant official emergency instruction, while avoiding posts that impersonate authorities. If an authority changes its advice, follow the current instruction rather than clinging to an earlier screenshot.
For example, “a service outage is being investigated” does not mean “drinking water is contaminated”, and “an image is unverified” does not mean “no emergency exists”. Precision prevents confusion.
Five warning signs of a manipulated crisis post
- No original source: many reposts, nobody can identify the first credible witness.
- Pressure to act without checking: demands to travel, pay or share urgently before confirmation.
- A familiar name with the wrong destination: an official-looking logo paired with a suspicious website or account.
- Inconsistent place or date: landmarks, weather or earlier postings do not support the caption.
- One video presented as universal proof: a short clip is used to claim an event is occurring everywhere.
None is definitive by itself. Together they are a reason to slow down, verify and avoid amplifying the claim.
The bottom line
AI-generated disinformation can make a crisis harder to navigate because convincing-looking evidence is no longer difficult to fabricate. The danger is not limited to fake images; authentic material can be miscaptioned, and cloned audio can exploit trust within families and organisations.
When a claim could affect your safety, verify the source and the action—not just the pixels. A direct official channel, independent confirmation and an offline family information plan are more reliable than a confident viral post.
Frequently asked questions
How can I tell if a crisis video was made by AI?
Start by locating the earliest source, checking place and date, and comparing the claim with independent evidence. Visual mistakes and automated detectors can provide clues but are not reliable verdicts.
Can real footage still be misinformation?
Yes. Authentic images may be old, filmed elsewhere or edited into a misleading context. Verifying the claim matters as much as authenticating the file.
Should I ignore an emergency message if I cannot verify it?
Do not automatically dismiss possible danger. Check trusted emergency channels and act on hazards you can directly observe or verified official instructions. Avoid sharing unverified directions as facts.
What if a relative's voice asks me to send money urgently?
Use an independent method such as a return call to their saved number or another trusted contact. Do not rely on a familiar voice alone.
What is the best backup if the internet fails?
A paper list of important contacts and an appropriate battery or wind-up radio can help, alongside an agreed family meeting and contact plan.
Sources and further reading
- Full Fact — A Fact-Checker’s Guide to Spotting AI-Generated Content (2025) — practical verification methods.
- Bellingcat — Online Investigation Toolkit — professional verification methodology.
- International AI Safety Report 2026 — synthetic content, manipulation and documented harm.
- British Red Cross — Five Tips to Plan for Emergencies — making trusted information part of household readiness.
- UK Government — Power Cuts — alternate communication and service information.
Research reviewed 6 November 2026. The water-warning example and infrastructure-disruption scenario are fictional training exercises.