#FactCheck -AI-Generated Video Falsely Claims Indian Soldiers Failed Motorcycle Stunts During Republic Day Parade
Executive Summary
As India concluded its 77th Republic Day celebrations on January 26, 2026, with grandeur and patriotic enthusiasm along the iconic Kartavya Path, a video began circulating on social media claiming to show Indian security personnel failing to perform motorcycle stunts during the ceremonial parade. The short clip allegedly depicts soldiers attempting high-risk, synchronised motorcycle manoeuvres, only to lose balance and fall off their bikes. The visuals were widely shared online with mocking captions, suggesting incompetence during a nationally televised event. However, an research by the CyberPeace found that the video is not authentic and was digitally generated using artificial intelligence.
Claim
A Pakistan-based X user, Sadaf Baloch (@sadafzbaloch), shared the video on January 27, claiming it showed Indian security personnel failing to execute motorcycle stunts during the Republic Day parade held on January 26, 2026. While sharing the clip, the user wrote:“Every time the Indian Army tries a tactical stunt, it looks less like combat training and more like a low-budget circus trailer filmed in one take.”The post was widely circulated with similar narratives questioning the professionalism of Indian forces.
Here is the link and archive link to the post, along with a screenshot.
To verify the authenticity of the viral video, the Desk conducted a detailed frame-by-frame analysis. During the examination, a watermark linked to ‘Sora’—an AI text-to-video generation model was detected at the 00:05 timestamp. The presence of this watermark strongly indicated that the video was artificially generated and not recorded during a real-world event.
Fact Check:
Further visual scrutiny revealed several inconsistencies commonly associated with AI-generated content. The background appeared unnatural and lacked realistic depth, while the movements and reactions of the security personnel looked mechanically exaggerated and inconsistent with real physics. Facial expressions and body motions during the alleged falls also appeared unrealistic. To strengthen the verification, the Desk analysed the clip using Sightengine, an AI-detection tool. The results showed a 98 per cent probability that the video contained AI-generated or deepfake elements.
Below is a screenshot of the result.
As part of the research , the Desk also conducted a customised keyword search and reviewed official coverage of the Republic Day parade. A full-length video broadcast by DD News on its official YouTube channel was examined. The footage showed joint CRPF and SSB motorcycle teams performing traditional daredevil stunts without any mishap. No incident resembling the viral claim was found in the official broadcast or in any credible media reports.
Here is the video link and a screenshot.
Conclusion
The CyberPeace research confirms that the viral video purportedly showing Indian security personnel failing to perform motorcycle stunts during the 77th Republic Day parade is AI-generated. The clip has been falsely circulated online as genuine content with the intent to mislead viewers and spread misinformation.
Related Blogs
Introduction
Digital evidence has become part of almost every modern investigation. A photograph can place a person at a location, an audio recording can capture a conversation, and a video can appear to show an event as it happened. For years, the main forensic concern was whether such material had been altered. The rapid growth of generative artificial intelligence has added a harder question: even when a file is preserved exactly as received, can investigators still trust what it appears to show?
Deepfakes have made this question practical rather than theoretical. Synthetic or manipulated audio, video and images can imitate real people and real events with increasing realism. CERT-In describes deepfakes as a high-risk threat because they can support disinformation, fraud, social engineering and reputational harm.[1] NIST research likewise treats AI-generated media as a digital-forensics challenge that requires systematic evaluation of detection technologies.[2]
The result is an evidence problem. The answer is not to stop trusting digital evidence, but to become more disciplined about establishing its origin, integrity, context and authenticity.
The evidence problem begins before the laboratory
When a suspicious video reaches an investigator through WhatsApp, Telegram, email or social media, the file may already have passed through several transformations. It may have been compressed, re-encoded, cropped, renamed or stripped of metadata. A screenshot may preserve what is visible but lose the original file structure. A forwarded audio clip may contain no reliable information about where it was first recorded.
For that reason, forensic examination should begin with acquisition and provenance, not with a quick “deepfake detector” result. Investigators should ask: Who supplied the file? Where was it obtained? Is there an original version? What device or account produced it? What happened to the file before it reached the investigator?
Cryptographic hashing remains important because it can demonstrate that an acquired working copy has not changed during examination. But a valid hash does not prove that the underlying event was genuine. A perfectly preserved fake is still a fake.
What a professional examination should look for
A reliable assessment combines several forms of evidence rather than relying on one technical indicator.
Source and acquisition. The original artefact should be preserved whenever possible. Investigators should record the acquisition method, date and time, source account or device, and any known transformations before collection. A documented chain of custody is essential when material may later support a legal, disciplinary or regulatory decision.
Metadata and file structure. Metadata may provide useful clues about creation, encoding, editing software and timestamps. File structure, compression behaviour and related technical characteristics can also reveal inconsistencies. However, these indicators are supporting evidence, not proof on their own, because metadata can be removed or rewritten during normal processing.
Content-level examination. Forensic analysis can include frame-by-frame video review, audio waveform and spectral examination, and inspection for inconsistencies in lighting, reflections, facial movement, lip synchronisation or background elements. Such signs may help guide an investigation, but they are not a permanent checklist. Generative systems continue to improve.
Independent corroboration. This is often the strongest step. If a recording allegedly shows that a person was in a particular place at a particular time, investigators can compare it with CCTV, access-control records, device artefacts, communications, location information, eyewitness accounts or other independent records. The goal is to determine whether the wider evidence supports the event represented by the media.
A real-world lesson: the Pikesville case
The 2024 Pikesville High School incident in Maryland provides a practical example of why authenticity cannot be assumed from appearance alone. An audio recording circulated online that was presented as the principal making racist and antisemitic comments. On January 17, 2024, Baltimore County Public Schools said it could not yet confirm the recording’s veracity and opened an investigation.[3]
Several months later, the school district reported that investigators, with assistance from the FBI and other experts, had verified that the audio had been created using artificial intelligence.[4] Police subsequently arrested the school’s former athletic director in connection with the fabricated recording.[5]
The forensic lesson is larger than the incident itself. The recording had social consequences before its authenticity was established. In a fast-moving online environment, the first version of an event can travel much further than the later correction. Deepfake investigations therefore have to consider not only whether media is authentic, but also how quickly unverified material can influence decisions.
From deepfake detection to content provenance
Detection tools will remain useful, but they should be treated as part of an examination rather than an automatic verdict. NIST’s Guardians of Forensic Evidence work reflects the need to evaluate how analytic systems perform against changing forms of AI-generated media and how well they generalise beyond controlled conditions.[2]
Another important direction is content provenance. The Coalition for Content Provenance and Authenticity (C2PA) has developed a technical framework for recording verifiable information about how digital content was created and changed. Content Credentials can bind provenance information to an asset using cryptographic techniques, allowing later users to inspect a recorded content history when that information is available.[6]
Provenance does not mean that every claim associated with a file is automatically true. It adds context: who created it, what actions were taken and how the asset changed. In a deepfake environment, that context can be as important as the content itself.
Why this matters in India
The issue is especially relevant to India’s fast-growing digital environment. CERT-In’s 2024 advisory identifies misinformation, fraud and reputational damage among the risks associated with synthetic media.[1] In August 2026, the Government of India stated that the regulatory framework addresses AI-generated deepfakes and noted amendments to the IT Rules in February 2026 concerning harms arising from synthetically generated information, including requirements related to labelling and traceable metadata for permissible AI-generated content.[7]
For organisations, deepfake response should therefore not be treated only as a media or public-relations issue. It can become an incident-response and forensic issue. A suspicious executive voice note, a manipulated employee video or a fabricated screen recording may require preservation, technical examination and independent corroboration before any action is taken.
Conclusion
Deepfakes do not make digital evidence useless. Deepfakes make handling of evidence more dangerous. The professional response is not to believe everything or to doubt everything. The professional response is to build a process around evidence: preserve the original where possible document how evidence was acquired, calculate and record hashes, examine metadata and technical characteristics use detection tools while understanding their limitations compare media with independent evidence and examine provenance information where it is available.
Importantly investigators and decision-makers should separate three questions: Is the file intact? Is the content authentic? Does the content actually prove the event being alleged? Deepfakes can pass the test while failing the other two.
In the age of AI evidence will increasingly be judged not only by how convincing it looks but, by how well its origin, integrity, context and history can be demonstrated. That is the standard that can help preserve trust when seeing and hearing're no longer enough.
References
1. CERT-In, “Deepfakes - Threats and Countermeasures,” Advisory CIAD-2024-0060, 27 November 2024. View source
2. NIST, “Guardians of Forensic Evidence: Evaluating Analytic Systems Against AI-Generated Deepfakes,” 27 January 2025. View source
3. Baltimore County Public Schools, “January 17, 2024, Community Update: Message from Superintendent Dr. Myriam Rogers Regarding Pikesville High School.” View source
4. Baltimore County Public Schools, “April 24, 2024 Staff and Community Update: Message from Superintendent Dr. Myriam Rogers Regarding Pikesville High School Investigation.” View source
5. The Baltimore Banner / WYPR, “Ex-athletic director framed principal with AI-generated voice, police say,” 25 April 2024. View source
6. Coalition for Content Provenance and Authenticity (C2PA), “Content Credentials: C2PA Technical Specification,” Version 2.1. View source
7. Government of India, Ministry of Electronics & Information Technology, “Government Strengthens Regulatory Framework to Address AI-Generated Deepfakes,” 6 August 2026. View source
Executive Summary
A post is being widely circulated on social media claiming that the entire Delhi airport was submerged following heavy rainfall. The post questions the airport’s drainage system and attributes the situation to the negligence of its staff. The post further claims that the airport’s arrangements were exposed after just one spell of rain and that its drainage system needs to be upgraded using better technology. A research by the research wing of CyberPeace found that the viral video is not from Delhi airport but dates back to April 2024. At the time, heavy rainfall in Dubai caused flood-like conditions at the airport, affecting flight operations on waterlogged runways as well. Thus, an old video from Dubai is being falsely shared on social media as footage of Delhi airport.
Claim:
A user on the social media platform X (formerly Twitter) shared the viral video with the caption: “Delhi Airport — the entire airport is submerged. This is the situation after just one spell of rain. A minor issue of drainage and negligence by the staff — the public is aware — now the negligence will be punished. ‘The drainage system technology must be developed.’”
https://x.com/RJMediaGroups/status/2096470339060736149
Factcheck
As part of our investigation, we conducted a reverse image search of keyframes from the video. During the search, we found the original video on a YouTube channel named The Indian Express, which was uploaded on April 17, 2024. According to the video description, the incident took place in Dubai. The post link and screenshot are provided below.
https://www.youtube.com/watch?v=O9F1qTMIO5M
During our investigation, we found visuals similar to those in the viral video on the YouTube channel AlArabiyaEnglish. The video was uploaded on April 16, 2024. According to the information provided in the video description, several parts of the UAE were affected by flooding following sudden heavy rainfall on April 16, 2024. The heavy rains also affected operations at Dubai Airport, disrupting several flights. According to the report, the UAE recorded a record 254 mm of rainfall in less than 24 hours. In view of the adverse weather conditions, Dubai Airport advised passengers to travel only if absolutely necessary. The post link and screenshot are provided below.
https://www.youtube.com/shorts/KzIstkiOdp4
Conclusion:
Our investigation found that the viral video does not show Delhi Airport after heavy rainfall. The visuals are actually from Dubai Airport, where severe rainfall in the UAE in April 2024 led to flooding and disruptions to airport operations. Therefore, the viral video is old and has been falsely shared on social media with the misleading claim that it shows the current condition of Delhi Airport.
Introduction
India’s new Policy for Data Sharing from the National Transport Repository (NTR) released by the Ministry of Road Transport and Highways (MoRTH) in August, 2025, can be seen as a constitutional turning point and a milestone in administrative efficiency. The state has established an unprecedentedly large unified infrastructure by combining the records of 390 million vehicles, 220 million driver’s licenses, and the streams from the e-challan, e-DAR, and FASTag systems. Its supporters hail its promise of private-sector innovation, data-driven research, and smooth governance. However, there is a troubling paradox beneath this facade of advancement: the very structures intended to improve citizen mobility may simultaneously strengthen widespread surveillance. Without strict protections, the NTR runs the risk of violating the constitutional trifecta of need, proportionality, and legality as stated in Puttaswamy v. UOI, which brings to light important issues at the nexus of liberty, law, and data.
The other pertinent question to be addressed is as India unifies one of its comprehensive datasets on citizen mobility the question becomes more pressing: while motorised citizens are now in the spotlight for accountability, what about the millions of other datasets that are still dispersed, unregulated, and shared inconsistently in the areas of health, education, telecom, and welfare?
The Legal Backdrop
MoRTH grounds its new policy in Sections 25A and 62B of the Motor Vehicles Act, 1988. Data is consolidated into a single repository since states are required by Section 136A to electronically monitor road safety. According to the policy, it complies with the Digital Personal Data Protection Act, 2023.
The DPDP Act itself, however, is rife with state exclusions, particularly Sections 7 and 17, which give government organisations access to personal information for “any function under any law” or for law enforcement purposes. This is where the constitutional issue lies. Prior judicial supervision, warrants, or independent checks are not necessary. With legislative approval, MoRTH is essentially creating a national vehicle database without any constitutional protections.
Data, Domination and the New Privacy Paradigm
As an efficiency and governance reform, VAHAN, SARATHI, e-challan, eDAR, and FASTag are being consolidated into a single National Transport Repository (NTR). However, centralising extensive mobility and identity-linked records on a large scale is more than just a technical advancement; it also changes how the state and private life interact. The NTR must therefore be interpreted through a more comprehensive privacy paradigm, one that acknowledges that data aggregation is a means of enhancing administrative capacity and has the potential to develop into a long-lasting tool of social control and surveillance unless both technological and constitutional restrictions are placed at the same time.
Two recent doctrinal developments sharpen this concern. First, the Supreme Court’s foundational ruling that privacy is a fundamental right remains the constitutional lodestar, any state interference must satisfy legality, necessity and proportionality (KS Puttaswamy & Anr. vs UOI). Second, as seen by the court’s most recent refusals to normalise ongoing, warrantless location monitoring, such as the ruling overturning bail requirements that required accused individuals to provide a Google maps pin, as movement tracking necessitates closer examination (Frank Vitus v. Narcotics Control Bureau & Ors.,).When taken as a whole, these authorities maintain that unrestricted, ongoing access to mobility and toll-transaction records is a constitutional issue and cannot be handled as an administrative convenience.
Structural Fault Lines in the NTR Framework
Fundamentally, the NTR policy generates structural vulnerabilities by providing nearly unrestricted access through APIs and even mass transfers on physical media to a broad range of parties, including insurance companies, law enforcement, and intelligence services. This design undermines constitutional protections in three ways: first, it makes it possible to draw conclusions about private life patterns that the Supreme Court has identified as one of the most sensitive data categories by exposing rich mobility trails like FASTag logs and vehicle-linked identities; Second, it allows bulk datasets to circulate outside the ministry’s custodial boundary, which creates the possibility of function creep, secondary use, and monetisation risks reminiscent of the bulk sharing regime that the government itself once abandoned; and third, it introduces coercive exclusion by tying private sector access to Aadhaar-based OTP consent.
Reference
