#FactCheck-Claim that a Chinese helicopter reached the site of a glacier collapse in Nepal is false; video is from Alaska

How clothing patterns, movement and real-world conditions can affect computer vision

Introduction

‍

A person walks in front of an AI camera. The person is clearly visible to us, yet the system may not always identify the person with the same confidence. This sounds strange until we understand one basic fact: a camera records an image, while an AI model interprets that image through patterns learned during training.

This idea became popular through demonstrations such as “The T-Shirt Invisibility Cloak,” where specially designed clothing was shown as a way to interfere with person-detection systems. The wearer does not become physically invisible. The camera still captures the person. Instead, the visual pattern can make a particular AI model less confident or cause an incorrect prediction.

So, how can a piece of clothing affect machine vision? And what does this tell us about the strengths and limitations of AI surveillance?

‍

How Does an AI Camera See?

‍

A normal CCTV camera mainly captures and records video. An AI camera adds software that analyses the video for a specific task, such as detecting a person, vehicle, face or event. 

A simple way to understand the process is:

‍

‍

The model does not understand a person exactly as a human does. It processes numerical representations learned from training data. For a person detector, the question is closer to “Do these visual patterns match the class ‘person’?” than “I know this is a human.”

‍

Think About It

‍
A human sees a person wearing a complex shirt and easily separates the shirt design from the person. An AI model must decide what the visual patterns in the image mean for its particular task.

‍

When Clothing Becomes Difficult

‍

‍

Clothing can become challenging when it changes the visual information available to a model.

Complex patterns can add many edges, repeated shapes and textures to an image. These do not automatically confuse an AI camera, but they can influence the features a detector uses.

Low contrast can create another problem. If clothing is close in colour or brightness to the background, the boundary between the person and the environment may be harder to separate, especially in poor lighting or low-resolution footage.

Movement creates another challenge. A loose jacket, flowing fabric or layered clothing changes shape as the person walks or turns. Researchers designing adversarial T-shirts have had to account for these non-rigid changes because a printed pattern does not stay flat on a moving body.

‍

The T-Shirt Invisibility Cloak

‍

‍

The video “The T-Shirt Invisibility Cloak” is a simple introduction to a larger research area called physical adversarial examples.

In adversarial machine learning, researchers study inputs deliberately designed to make a model produce an incorrect prediction. With clothing, the visual pattern is physically printed on a garment and then captured by a camera.

A 2019 study by Xu and colleagues demonstrated an adversarial T-shirt designed to interfere with person detectors in the physical world. Under their test conditions, the reported physical attack success rate against YOLOv2 was 57%. The work also modelled cloth deformation caused by movement.

A 2022 CVPR study on adversarial texture extended the idea to different viewing angles and tested physical clothing such as T-shirts, skirts and dresses.

But there is an important limitation: an adversarial garment is not a universal invisibility cloak. Different cameras and AI systems use different models, training data and processing pipelines. A pattern that affects one detector may have little effect on another.

‍

Why Does the AI Get It Wrong?

‍

Clothing may be only one part of the problem. A model can receive features different from what it learned during training. Camera angle, distance, movement, lighting, motion blur, compression and occlusion can also change the image.

Imagine the same person in four situations:

‍

‍

There may not be one fixed answer. AI performance depends on the combination of conditions.

‍

AI Camera vs Human Vision

‍

Humans use context. We understand that clothes wrinkle, people turn, shadows change and objects may temporarily block part of a body. An AI detector is more task-specific and can become less reliable when input conditions differ from the data used to train or evaluate it.

This does not make AI useless. It makes realistic testing important.

‍

Building More Reliable AI Surveillance

‍

The lesson is not that AI cameras should be abandoned. They should be tested under difficult, realistic conditions.

Testing can include different clothing patterns, colours, body positions, distances, camera angles, lighting conditions and crowded scenes. Security teams should also distinguish between person detection, facial recognition, tracking and behaviour analysis because these are different tasks. ^[6]

Most importantly, an automated alert should be treated as a prediction, not unquestionable truth. Important decisions should include appropriate human review.

‍

A New Research Direction

‍

Research is also moving beyond ordinary visible-light cameras. A 2026 CVPR paper explored adversarial clothing designed to affect both visible and infrared surveillance systems. At the same time, researchers continue to study ways of making models more robust against physical adversarial attacks.

This creates a continuing security cycle:

‍

‍

That cycle is a normal part of security research.

‍

‍

At a Glance: Human Vision vs AI Detection

‍

‍

Conclusion

‍

The idea of an “invisible T-shirt” is fascinating because it challenges a common assumption about AI: if a machine can see an image, we expect it to understand that image correctly.

A person can remain completely visible while a particular AI detector becomes less confident or makes a wrong prediction. Clothing patterns, texture, folds, movement, lighting, camera angle and other factors can interact with a model in unexpected ways. Research has demonstrated physical adversarial clothing against person detectors, while newer work is exploring more complex surveillance conditions.

The real lesson is not that a shirt can make someone invisible. AI vision is a prediction system with strengths and weaknesses. Understanding those weaknesses helps researchers build better defences, helps security teams evaluate systems realistically, and reminds us that automated surveillance should be deployed with technical care, human oversight and respect for privacy.

‍

References

• The T-Shirt Invisibility Cloak
• YouTube video referenced for the article’s introductory example: https://www.youtube.com/watch?v=NyofHyRm5CQ
• Xu, K., Zhang, G., Liu, S., Fan, Q., Sun, M., Chen, H., Chen, P.-Y., Wang, Y., & Lin, X. (2020). Adversarial T-shirt! Evading Person Detectors in a Physical World. Computer Vision – ECCV 2020, 665–681. https://doi.org/10.1007/978-3-030-58558-7_39
• Study demonstrating a physical adversarial T-shirt against person detection and modelling non-rigid cloth deformation during movement. https://arxiv.org/abs/1910.11099
• Thys, S., Van Ranst, W., & Goedemé, T. (2019). Fooling Automated Surveillance Cameras: Adversarial Patches to Attack Person Detection. CVPR Workshops.
• Earlier work showing physical adversarial patches can reduce the accuracy of person detectors. https://openaccess.thecvf.com/content_CVPRW_2019/html/CV-COPS/Thys_Fooling_Automated_Surveillance_Cameras_Adversarial_Patches_to_Attack_Person_Detection_CVPRW_2019_paper.html
• Hu, Z., Huang, S., Zhu, X., Sun, F., Zhang, B., & Hu, X. (2022). Adversarial Texture for Fooling Person Detectors in the Physical World. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 13307–13316.
• Study extending physical adversarial clothing to multiple viewing angles and garments including T-shirts, skirts and dresses. https://openaccess.thecvf.com/content/CVPR2022/html/Hu_Adversarial_Texture_for_Fooling_Person_Detectors_in_the_Physical_World_CVPR_2022_paper.html
• Long, J., Jiang, T., Liu, H., Ma, C., Zhou, W., Yang, Y., & Yao, W. (2026). Thermally Activated Dual-Modal Adversarial Clothing against AI Surveillance Systems. CVPR 2026.
• Recent work exploring adversarial clothing against visible and infrared surveillance systems. https://openaccess.thecvf.com/content/CVPR2026/html/Long_Thermally_Activated_Dual-Modal_Adversarial_Clothing_against_AI_Surveillance_Systems_CVPR_2026_paper.html
• Li, S., Zhang, S., Chen, G., Wang, D., Feng, P., Wang, J., Liu, A., Yi, X., & Liu, X. (2023). Towards Benchmarking and Assessing Visual Naturalness of Physical World Adversarial Attacks. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 12324–12333.
• Useful context on physical adversarial attacks and the challenge of making them visually natural. https://openaccess.thecvf.com/content/CVPR2023/papers/Li_Towards_Benchmarking_and_Assessing_Visual_Naturalness_of_Physical_World_Adversarial_CVPR_2023_paper.pdf ^[6]

‍

‍

Executive Summary

A viral image circulating on social media claims that Tamil Nadu Chief Minister C. Joseph Vijay touched the feet of Congress leader Rahul Gandhi during his swearing-in ceremony, while Congress and several other parties extended support to his government. The image is being widely shared with captions suggesting it captures a real political moment. However, CyberPeace Research Wing research has found the claim to be false. The image is AI-generated and does not depict any real event.

‍

Claim

‍

A Facebook user shared the viral image on May 10, 2026, claiming that TVK chief and actor Vijay had taken oath as the Chief Minister of Tamil Nadu. The post further claimed that during the ceremony, Vijay touched Rahul Gandhi’s feet to seek blessings, and the gesture was applauded by leaders present on stage. The post, along with archived links and screenshots, is being circulated as authentic evidence of the alleged incident.

• https://www.facebook.com/100057774695228/posts/1389222123013598/?rdid=FEzRYpVvSIieeUbj# 
• https://archive.ph/kv4e1 

‍

‍

‍

Fact Check
‍

A keyword-based search on Google did not return any credible news reports supporting the claim or confirming such an event. A closer visual examination of the image raised strong suspicions of AI manipulation, prompting verification through AI detection tools. When the image was analyzed using the SIGHTENGINE detection tool, the results indicated that the image is 99% likely to be AI-generated.

‍

‍

Further verification using another AI detection platform, HIVE MODERATION, also flagged the image as synthetic, showing an 81% probability of being AI-generated.

‍

‍

Conclusion

‍

The research clearly shows that the viral image is not real. It has been generated using artificial intelligence and is being falsely shared as a real political event.

‍

Introduction

It might seem too good to be true: free movies, live TV and sports without any subscription plan are deals that no one in India could actually refuse, and that’s precisely the explanation as to why applications such as Pikashow have gained tremendous traction in India and all over the world as well. However, cybersecurity authorities have a strong cause of concern: the threat that this type of application poses towards lakhs of Indians with malicious software, stealing information and even money. Pikashow may be a perfect case study to establish this problem, but in fact, it all ties down to how downloading anything without the recognised applications has certain risks attached to it.

The Pikashow Warning

With no monthly subscription fees, Pikashow is a popular free app for streaming movies, web series, live TV channels and sports programmes. Since its development, Pikashow has had a reputation for never being found on the Google Play Store or Apple’s App Store, which necessitates a process of side-loading. A side load, in this case, entails downloading the Pikashow APK file from an unofficial third-party source and manually installing the app, circumventing the security measures of official download channels.

The latter is significant because security experts who have analysed Pikashow argue that it acts as an informal content aggregator; movies and TV series are illegally scooped up from online streaming services, then shared via various third-party links and streamed without a content licence or any agreement with legitimate providers.

Without going through the rigorous checks conducted by Google or Apple, security researchers discovered that even while the app appeared to be working fine with its promised streams, hidden malware within the installer can run silently, leading to potentially significant security risks such as data theft of user logins, one-time passwords, and private files, including banking credentials. This has been cited to be so problematic for India that law enforcement agencies like the Indian Cyber Crime Coordination Centre under the Ministry of Home Affairs and the Cyber Dost platform are expected to have issued public advisories linking Pikashow to a proliferation of cases of cybercrime. Users should take into account the legal implications alongside security concerns, as using piracy apps may not only lead to device-compromising issues but can also be against the Copyright Act of India as well, thus increasing the risk of legal prosecution.

Why This Isn't Just a "Pikashow Problem"

This should not be viewed as a single warning concerning one app. Pikashow is merely an illustration of one app from millions that exists outside official app stores and is built with the same mechanics that make Pikashow dangerous and applicable to virtually any app, game, or "modded" file which is downloaded from a random website, third-party app store, or Telegram channel.
Some issues that persist are- 

1. Official app stores are like a filter that is not perfect, but it is still critical to the security on a user's device. Both Google and Apple's App stores run automatic and manual malware scanners, screen permissions for legitimate use, and have the ability to remove malicious developers as necessary. Empirical research has estimated just how important this is: more than fifty-times more malware exists on sideloaded, "unofficial" apps versus official. Mobile security firm Zimperium claims nearly 40% of all devices tested with malware were infected due to downloading an app from outside the official app stores; furthermore, nearly 80% of malicious, sideloaded apps contained riskware or trojans.
2. Informational channels can also be used to gain the trust of users by mimicking familiar brands One frequently mentioned scheme is one where malicious actors repackage popular apps with a malignant intent to syphon the app, disguised as a free, trustworthy app. Often when these apps are available, they work exactly the same as the legitimate app, as a matter of fact, but quietly in the background steal passwords or log keystrokes; this was discovered to be occurring with a financial trojan called TeaBot, where bank details and login info were intercepted through the device's Accessibility Service.
3. There is generally no recourse or legal liability that applies once something has gone wrong. While official app stores are official developers that can be identified to submit complaints against them, if a third-party download leads to damages, that developer generally cannot be located to establish liability or legal standing.
4. Excessive app permissions will often go undetected This occurs for the similar reasons that a malicious actor can include such a clause as explained above. There's really nothing to stop the developer from asking to utilise one's SMS, call log, camera, microphone, storage or location when there's not one reason for one application to need access to everything on the device.
5. Expired software has the same or more risk It's actually very common to see apps that receive security updates out in the wild that either don't have the appropriate update sent over to third-party app stores or third-party stores don't have the capacity to update the affected applications. That's why most official third-party app downloads should be of fairly new versions; for example, outdated versions of a banking app, which is being patched as well as receiving other updates from Google or Apple still, will continue to persist on third-party download sites for some time after patching, which could then be malicious.
6. Even legitimate, globally recognised apps can and have been compromised The same vulnerability was noted for TikTok in 2023, which, when found on unofficial APK download sites, contained hidden data-tracking malware yet looked identical to the official app version.

How to Protect Yourself

The reassuring thing is that the vast majority of this risk can be eliminated through simple, repeatable habits. Use trusted sources like Google Play and Apple's App Store, which are not faultless, of course, but they remove many infected apps from circulation following review.

1. Be sceptical of "free" premium-paid apps.
2. If something in a file from a site you don't recognise offers content or features that are normally behind a paywall and they want nothing to do with it, that is a bad sign, not a good deal. Review app permissions when you go to install anything. A flashlight doesn't need access to your SMS messages or contacts, as one recently downloaded on a colleague's Android allegedly attempted to acquire them from the Google Play store. After sideloading any app that you do not fully trust, it's always a wise idea to uninstall it and scan the handset to ensure malicious files aren't still lingering around.
3. Choose legitimate services to avoid malware.
4. In entertainment terms this will apply most clearly to piracy but applies similarly elsewhere in many senses. Trusted sources that are licensed include YouTube, JioCinema, Hotstar, Netflix, Amazon Prime Video, and MX Player, among others, which will offer content similar to illegally acquired or pirated versions without the security, as well as the legal risks. If you have recently used the file installation route to any degree of caution on a sensitive device, then also monitor banking and payment apps.
5. That is not least because the most reported consequence of infection from an illicit or unknown source is the theft of financial account login details.

Conclusion

Free entertainment can look tempting, but as cybersecurity experts keep pointing out, "free" often just means the cost has been shifted from your wallet to your data, your device, and potentially your bank account. The safest rule of thumb remains simple: if an app isn't on an official store, ask why and think twice before installing it.

Sources

• Pikashow app risky, may expose users to malware and data theft: Experts — Times of India
• Free OTT Hack or Cyber Trap? Pikashow Flagged for Serious Security Risks — The420.in
• Pikashow Warning: Free Movies Could Empty Your Bank Account, Government Issues Strong Alert — The420.in
• Is Pikashow Safe or Risky to Use in 2025? Expert's Advice — AiPlex AntiPiracy
• Why Pikashow May Not Be Safe for Your Device — AiPlex AntiPiracy
• Concerns Rise Over the Safety of Popular Streaming App Pikashow — openPR
• Beyond the App Store: The Hidden Risks of Sideloading Apps — Zimperium
• App Sideloading: Risks, Rules, and How IT Admins Respond — Trio
• What Are the Risks of Sideloading Apps on Your Smartphone? — Bitdefender
• Sideloading Risk: Alternative App Stores and Brand Protection — Allure Security
• Building a Trusted Ecosystem for Millions of Apps: A Threat Analysis of Sideloading — Apple

‍