AI-edited image in bdnews24.com’s report on tourist’s stolen bag
bdnews24.com published a report identifying a young man from CCTV footage in connection with the theft of a German tourist’s bag containing a laptop, a camera and other items at Natore Railway Station. However, the image used as the report’s cover was edited using AI technology. Several visible elements in the image, including the tourist’s face, were altered. Neither the image caption nor any other part of the report disclosed that the image had been edited using AI.
According to an October 8 report (archived version) by bdnews24.com, Krill Lander, an 18-year-old German citizen, was staying in the waiting room at Natore Railway Station while travelling in Bangladesh, before heading to India. At that time, a young man aged between 20 and 25 struck up a friendship with him and later fled with his bag. The tourist filed a general diary (GD), a police record of an incident, at Natore Sadar Police Station.
According to the officer-in-charge of Santahar Railway Police Station, as cited in the report, police collected and reviewed CCTV footage from Natore Railway Station. The report states that footage recorded at 3:39 p.m. on Wednesday showed a young man wearing a yellow coat calling the German tourist over and leading him away. Sixteen minutes later, the same young man was seen removing his coat and leaving with a bag, according to the report.
Although the report used images taken from the original CCTV footage, several inconsistencies were observed in the cover image. An object visible near the German tourist’s left hand in the original footage was missing from the cover image. The tourist’s face also appeared distorted in the cover image. The shirt pattern of the person wearing a yellow coat, who was suspected in the theft, had changed to a vertical check pattern. Although his face appeared much clearer in the cover image than in the original footage, it could not be determined with certainty whether the faces in the two images matched. These changes prompted further examination of how the image had been edited.
The image was examined using the SynthID detection tool for further verification. The result stated, “This media was made or edited with OpenAI.” This indicates that evidence was found that the media examined had been created or edited using OpenAI technology. The cover image is not a direct, unaltered representation of a frame from the original CCTV footage. The report provided no information or disclaimer indicating that the image had been edited using AI.
This distinction is important when CCTV footage is used as evidence, particularly in connection with a crime. There is a difference between edits such as increasing brightness or contrast and resizing an ordinary image or video, and using AI to recreate unclear areas. AI can fill in less visible parts of an image by generating new pixels or details based on assumptions. As a result, the image may contain new visual information beyond what was captured by the original camera.
Why AI-Edited Images Are Risky in Ongoing Investigations
A report published by The Washington Post on January 9 this year demonstrated the serious risks of using AI to “enhance” unclear images or videos. In the incident on January 7, after a person was killed by a gunshot fired by an Immigration and Customs Enforcement (ICE) officer in Minneapolis, attempts were made to use AI to “reveal” the officer’s face from a video that had circulated on social media. However, different AI-generated images based on the same video showed different faces, leading to the incorrect identification of a person as the officer.
The same problem emerged in the killing of Charlie Kirk, a conservative activist in the United States. Using AI to enhance blurry videos or images created confusion over the suspect’s appearance. Matt Moynihan, chief executive of AI detection firm GetReal Security, told The Washington Post, “AI tools can only reconstruct reality based on past information. But that is not actual reality. If you are not an AI expert, you are more likely to do harm than good.”
In Minneapolis, nurse Alex Pretti was killed by federal immigration officers on January 24. In that case, a “high-resolution” image was generated using AI from a blurry video. After viewing the image, many people claimed that the object in Pretti’s hand was a weapon. However, after examining the original video, AFP found that he was holding a phone. Hany Farid, a professor at the University of California, Berkeley, told AFP, “The problem with these images is that when you try to enhance an image using AI, it adds extra objects (hallucinations).”
A recent fact-check by Thai PBS Verify also demonstrated how dangerous it can be to use AI to clarify blurry images for investigative purposes. In July 2026, following an attack in Thailand’s Narathiwat province, a blurry image taken from CCTV footage was edited using AI and circulated on social media. The image appeared to show a weapon in the suspect’s hand and a tattoo on the hand. However, Thai PBS Verify examined the original CCTV footage and found that there was no tattoo.
Assistant Professor Dr. Jetsada Salathong of Chulalongkorn University explained that AI upscaling does not recover “hidden information” in blurry images. Instead, it relies on past information to redraw the image based on “statistical predictions” of what the unclear pixels might represent. As a result, an image may look clearer without being a faithful representation of the original.
