AI-Doctored Video Falsely Depicts Bhullar Resignation
On August 8, 2026, Newschecker and The Quint found that a viral video falsely depicts Gurpreet Singh Bhullar, the former Amritsar Police Commissioner, resigning and supporting the Khalistan movement. Reverse-image searches traced the underlying footage to an ANI report about police action against two ISI-backed terror modules, while multiple AI-audio detection tools flagged the altered soundtrack as likely synthetic or manipulated.
A viral video falsely claiming that Gurpreet Singh Bhullar, the former Amritsar Police Commissioner, resigned alongside other officers and endorsed the Khalistan movement has been digitally altered, according to separate fact checks by The Quint and Newschecker.
Both outlets traced the visual footage to an August 5 ANI video. In the original report, Bhullar discussed police action against two ISI-backed terror modules, including the arrest of nine people and the recovery of petrol bombs, pistols, a motorcycle, a lighter, and live cartridges. Neither fact check found Bhullar announcing a resignation or voicing support for the Khalistan movement in the source footage.
How the claim was checked
The Quint and Newschecker each report using reverse-image searches of video keyframes to identify the ANI footage. Newschecker also cited a government release stating that Harmanbir Singh, DIG Border Range, Amritsar, had replaced Bhullar as Commissioner of Police, Amritsar, while Bhullar was directed to report to the DGP Office until further orders.
The fabricated version overlays a different spoken message onto the authentic video. The Quint reports that its analysis with an unnamed AI-detection tool found a 99% likelihood of AI-generated speech, and that Hiya Deepfake Voice Detector also flagged the audio as AI-generated. Newschecker reports that Hive Moderation identified the media as likely AI-generated or manipulated, while Hiya Deepfake Voice Detector, Resemble AI, and Undetectable AI separately flagged the voice or audio as likely synthetic.
Detection results require corroboration
The strongest evidence in this case is the mismatch between the viral narration and the independently located original footage, rather than any one detector score. AI voice detectors can produce false positives and false negatives, particularly on compressed social-video audio. Here, however, the source-video comparison provides a direct verification path: the same imagery appeared in ANI's report, but its original audio concerned a terrorism investigation rather than resignation or political support.
For ML and trust-and-safety teams, the incident illustrates a common deepfake distribution pattern: real news footage supplies visual plausibility, while generated or manipulated audio introduces a high-impact false claim. Verification workflows that combine provenance checks, reverse-image search, original-source retrieval, and synthetic-media detection are generally more robust than reliance on a single classifier.
The Quint and Newschecker concluded that the viral video does not authentically show Bhullar resigning or endorsing the Khalistan movement.
Key Points
- 1Fact checks traced the clip's visuals to ANI footage about terror-module arrests, contradicting the viral resignation and political-support narration.
- 2Multiple detection tools flagged the altered audio, but source-footage provenance offers the clearest corroboration for the manipulation finding.
- 3Comparable misinformation campaigns often pair authentic video with synthetic speech, requiring provenance checks alongside deepfake classifier outputs.
Scoring Rationale
This is a concrete synthetic-media misinformation case involving a senior police official and a politically sensitive claim. It is relevant to trust-and-safety, media-forensics, and content-provenance practitioners, although it does not introduce a new model, tool, or broad platform policy.
Sources
Public references used for this report.
Practice interview problems based on real data
1,625 SQL & Python problems across 15 industry datasets — the exact type of data you work with.
Try 250 free problems
