Anatomy of a disaster-misinformation wave: the fake Assam flood videos

The story in 60 seconds

• A grim video presented as flood deaths in Guwahati was staged entertainment: BOOM traced the original to creator Simran SK on July 3, before the most destructive phase of the Upper Assam floods.
• A separate clip of people stranded on rooftops was synthetic. BOOM reported a 98.8% AI probability from Hive Moderation; Newschecker separately recorded scores of 99.2% from Hive, 100% from one Deepfake-o-Meter model and 99% from Sightengine.
• A collapsing shopping complex was real footage, but not from Assam. Newschecker and Fact Crescendo traced it to Beerwah in Budgam, Jammu and Kashmir, on July 21.
• A photograph of a man carrying two children through neck-deep water was online by August 2024 and linked by Fact Crescendo to Bangladesh—not the 2026 Assam disaster.
• The supplied brief described a Kerala video, but HeadlineDecoded found no original 2026 fact-check substantiating that attribution. We have not repeated it.

AI SummaryAI-generated summary, reviewed by editors

Four viral Assam flood claims reveal four misinformation techniques: staged performance, AI generation, false location and recycled imagery. Here is how fact-checkers exposed them.

▮▶ Switch to Shorts

Image note: The lead image is an AI-generated editorial illustration representing the verification process. It is not documentary evidence from the Assam floods.

Real disasters create ideal conditions for false context. People urgently want images, platforms reward dramatic material, and an old or artificial clip can travel across languages before anyone checks where it came from. The 2026 Assam floods produced a compact example of the problem: staged performance, generative video, genuine footage with the wrong location, and old imagery recaptioned as current.

These are not four versions of one hoax. They are four different routes by which visual evidence becomes misleading. That distinction matters because each route requires a different verification method.

1. The “dead bodies” were performers in a staged reel

One viral video showed several women lying motionless in moving water. Posts described the scene as the aftermath of severe flooding in Guwahati. The imagery was emotionally forceful, but its publication history contradicted that claim.

View the original: Simran SK's Instagram Reel (@simransk944).

BOOM’s July 29 fact-check, reported by Jagriti Trisha, reverse-searched frames and located the original on the Instagram and Facebook accounts of Simran SK. The creator had uploaded it on July 3 with laughing emoji. BOOM also reviewed the account’s broader output and found similar dramatic comedy performances; the profile identified Simran SK as a digital creator.

The date is decisive. The original performance was public weeks before the sharp deterioration in Upper Assam after July 18–19 described in contemporaneous reporting. The account history supplies a second, independent clue: the scene belongs to a recurring entertainment format, not eyewitness disaster documentation.

Verdict: staged and falsely contextualised. The people shown should not be described as flood fatalities.

2. The rooftop flood video was generated with AI

Another widely shared clip showed people sitting on roofs in an unnaturally uniform flooded settlement. Some posts claimed Assam had broken a 60-year flood record. Here, there was no earlier real-world location to recover because the scene itself showed signs of synthesis.

BOOM noted implausible water movement, artificial-looking ducks, repeated roof geometry and a house disappearing into the water in an unnatural way. It then tested the clip with Hive Moderation, which returned a 98.8% likelihood of AI generation.

Newschecker’s independent verification, written by JP Tripathi, reported similar visual faults and ran several detectors. It recorded 99.2% from Hive Moderation, 100% from the AVSRDD 2025 model available through Deepfake-o-Meter, and 99% from Sightengine.

Those percentages are tool outputs, not mathematical proof. AI detectors can produce false positives and should never stand alone. Here they reinforced observable anomalies and the absence of matching footage in credible reporting. The convergence of visual inspection, source searches and multiple detectors made the conclusion substantially stronger.

Verdict: AI-generated, not authentic footage of Assam.

3. The collapsing building was real—but it was in Jammu and Kashmir

A third video showed a multi-storey shopping complex being swept away by raging water. The event was real, which made the false Assam caption harder to spot.

Newschecker’s July 29 report, by Vasudha Beri and edited by Kushel Madhusoodan, reverse-searched video frames and found the same footage in a Greater Kashmir YouTube report dated July 21. Greater Kashmir located the collapse in Beerwah, Budgam district, Jammu and Kashmir. Newschecker also found a Times of India report using the same footage and reporting that floodwater weakened the building’s foundation; no casualties were reported.

Fact Crescendo separately published the same location finding. The important lesson is that “real video” does not mean “correct claim.” Reverse-searching keyframes and comparing architecture, terrain and publication dates can expose a location swap.

Verdict: authentic footage with a false location. It did not show the Assam floods.

4. The 2024 flood image was recycled as current

A photograph of a man carrying two children through neck-deep water was shared as a scene from the current Assam emergency.

Fact Crescendo’s July 27 investigation, by Debanjana Baishya, reverse-searched the image and found it in a Bangladeshi media report identifying the scene as flooding in Bangladesh. It also found matching material on Facebook from August 25, 2024, and YouTube from August 28, 2024.

The outlet’s conclusion contains an awkward sentence saying it “could verify” the exact location and time, while the surrounding evidence and final verdict say the precise details were not established. The defensible finding is narrower: the image was online by August 2024 and therefore cannot depict the 2026 Assam flood. The Bangladesh attribution comes from the earlier media report cited by Fact Crescendo.

Verdict: old and unrelated to the 2026 Assam floods.

A correction to the original tip

The initial story brief supplied to HeadlineDecoded claimed that a Kerala flood video had been relabelled as Assam. We searched for the underlying 2026 fact-check and did not find an original outlet report supporting that specific description. The verified, current location-swap case concerns Beerwah in Jammu and Kashmir. Publishing the Kerala claim merely because it appeared in a tip would reproduce the same failure this article examines.

What connects the four cases

All four exploit an evidence gap: the viewer sees something emotionally compatible with a flood but lacks the time, location or provenance needed to interpret it. The caption supplies that missing context—and can supply it falsely.

• Staging borrows the appearance of eyewitness footage.
• Generative AI creates an event that never occurred.
• Geographic relabelling moves a genuine event to a more topical location.
• Temporal relabelling moves old material into the present.

The common accelerant is urgency. During a disaster, users may share first because the scene feels representative, even when the exact claim is unverified. But “something like this is happening” is not evidence that this image shows it.

Practical checklist: verify before you amplify

1. Pause on high-emotion posts. Shock is a reason to check, not a reason to share.
2. Read the claim precisely. Identify the alleged place, date and event. A real flood video may still have the wrong caption.
3. Find the earliest upload. Search distinctive phrases and reverse-search several video keyframes, not only the thumbnail.
4. Inspect the account. Comedy, visual-effects and AI channels often disclose their format elsewhere even when a repost removes it.
5. Look for physical inconsistencies. Watch hands, faces, reflections, repeated buildings, moving water, vehicle geometry and objects that change between frames.
6. Use AI detectors cautiously. Treat scores as supporting signals. Seek agreement between visible anomalies, provenance searches and multiple tools.
7. Check local reporting. Search reputable outlets in the claimed location and compare landmarks, terrain, weather and official descriptions.
8. Open the source, not just a screenshot. Cropped reposts remove dates, captions and creator disclosures.
9. Preserve uncertainty. If you can only prove a clip predates the disaster, say exactly that. Do not invent a location to make the debunk sound complete.
10. Do not turn debunks into amplification. Label false material before displaying it and avoid looping disturbing imagery without a clear public-interest reason.

Sources and methodology

HeadlineDecoded reviewed the original fact-check reports from BOOM, Newschecker and Fact Crescendo linked above, including their described reverse-image searches, source uploads and detector results. We also compared the timing of the flood emergency with contemporaneous reporting. We did not treat the commissioning brief as evidence.

Abbreviations and full forms

• AI: Artificial Intelligence