Fake AI clips spread worldwide, each debunked a different way

Three fake AI clips were debunked in the last week, two of them deepfakes of named officials, and each one fell to a different check. The cases ran from the occupied West Bank to India, caught by a Google watermark, an army fact-checking unit and commercial AI detectors.

All three surfaced within days of one another, and none was flagged by the platform that carried it. Outside reviewers and detection tools did the work.

How the fakes were caught

The West Bank olive grove clip is fake. France 24's Observers team reported on October 23, 2026 that a video circulating since October 14, which claimed to show Israeli bulldozers uprooting olive trees in the occupied West Bank, was generated by artificial intelligence. The footage drew more than 200,000 views across several X posts, and it was shared by the Palestinian Authority's communications team before that post was deleted, according to the outlet.

The tell was a watermark. France 24 put the clip through SynthID, Google's detection tool, which found a mark showing the video came from Gemini, Google's chatbot. The clip's maker had pasted a thumbs-up image over the corner where Gemini's watermark usually sits, the outlet reported. France 24 also listed AI-style errors: a digger's shovel appears to sprout spikes before looking normal three seconds later, and a tree in the background falls before the machine reaches it.

The destruction of Palestinian olive groves is a documented problem. Reuters photographed Israeli Army bulldozers uprooting trees in Deir Nizam in late September 2026, and the UN says 286,000 olive trees were damaged in the West Bank between 2023 and the end of September 2026. This clip was not footage of that.

The Kalia video is fake. The Indian Army's fact-checking unit said on October 26, 2026 that a digitally altered clip carried fabricated remarks attributed to Lt Gen Girish Kalia, then a Major General. The fake claimed he admitted China had taken control of areas in Arunachal Pradesh and that troops had been denied supplies. The Army said the clip layered AI lip-syncing and voice cloning over real archival footage of a press briefing Kalia gave in Kupwara, Jammu and Kashmir, on October 27, 2023, where he reported 10 foiled infiltration attempts and the elimination of 27 militants. "A digitally altered video is being circulated with fabricated remarks attributed to Lt Gen Girish Kalia, claiming that China has taken control of Asaphila, Longjou and Taksing Circle in Arunachal Pradesh and that the Indian Army has been denied essential supplies," the unit said.

The Doval clip is fake. The Quint reported on October 26, 2026 that a video putting fabricated remarks about detained political leaders into the mouth of National Security Advisor Ajit Doval was a deepfake. A post carrying it had been viewed more than 380,000 times at the time of the outlet's report. The Quint traced the visuals to a YouTube video of Doval from March 19, 2019. Hive Moderation scored the clip at 99.4 percent likely to carry AI-generated audio, Hiya's voice detector rated the audio 3 out of 100, and Meta's AI detector showed the video had been generated with Meta's AI tools on October 25, 2026, The Quint reported.

The three cases show why no single check settles a clip. A watermark test works only when the maker leaves the mark in place, and this maker tried to cover it. For the other two, the giveaway was the footage itself: an archival briefing and a 2019 clip that did not match the new words.

SynthID's reach is widening. Google opened the SynthID Detector to anyone in English on October 7, 2026, after an early version for journalists and media professionals, and says it has watermarked more than 180 billion images and videos. The tool checks content made with Google's AI and partners including OpenAI, NVIDIA and Kakao.

Anyone can test a suspect clip at synthid.com, Google's free watermark portal, or upload it to Hive Moderation's detector.

Detection tools

• SynthID Detector: Google's watermark portal, open to everyone in English since October 7, 2026
• Hive Moderation: returns a confidence score for whether media is AI-generated
• Hiya: deepfake voice detector used by The Quint on the Doval clip