The Brief

Israeli Prime Minister Benjamin Netanyahu posted a video from a café today — holding up both hands to the camera, counting his fingers — after a two-week disinformation campaign involving AI-generated images of his death and a viral “six fingers” conspiracy flooded social media with over 145 million views. The campaign, traced to Iranian state-linked media and coordinated fake account networks, began after joint US-Israeli strikes killed Supreme Leader Khamenei on February 28 and escalated through fabricated images, a deepfake news presenter, and the misidentification of a press conference shadow as evidence of AI generation.

The Report

Netanyahu responded to weeks of death rumours on Saturday with a video posted to X from a coffee shop, captioning it “They say I’m what? Watch.” In the clip, he held up both hands one by one — a direct rebuttal to claims that a press conference video had shown him with six fingers, supposedly proving it was AI-generated. “I am dead for coffee,” he said, deploying a Hebrew idiom meaning he loves something intensely. His office had earlier issued a flat denial to Anadolu Agency: “These are fake news; the Prime Minister is fine.”

The disinformation arrived in two distinct waves. The first began within days of the February 28 strikes that killed Khamenei. AI-generated images — one showing Netanyahu bloodied in rubble, flagged at 93.5% confidence as synthetic; another depicting him surrounded by rescue workers, flagged at 76.9% — spread rapidly across X, Instagram, Telegram, and TikTok. A deepfake video of a fabricated news presenter claimed Israeli media had confirmed Netanyahu’s death, though it exhibited obvious lip-sync failures. Tasnim News Agency, which the US Treasury lists as linked to Iran’s Islamic Revolutionary Guard Corps, amplified the claims using circumstantial fragments: Netanyahu’s limited public appearances, a postponed diplomatic visit, an undated French readout of a Macron call.

Former US intelligence officer Scott Ritter, speaking on Russian state broadcaster RT, alleged that Iran had struck Netanyahu’s residence and killed his brother Iddo — claims that were then re-cited by Tasnim, creating a circular amplification loop between Russian and Iranian state media with no independent source at the base.

The second wave arrived after Netanyahu’s March 12 live press conference. Rather than accept the appearance as proof of life, social media users isolated a frame at the 0:35 mark where a shadow on his palm appeared to show a sixth finger — a known artefact of AI image generation. The claim spread through accounts including conservative commentator Candace Owens, who asked “Where’s Bibi?” and questioned why his office was “releasing and deleting fake AI videos.” A deepfake detection tool analysed the video and returned a 0.1% likelihood of AI generation. PolitiFact attributed the visual to “a trick of light” on the hypothenar eminence of his palm. X’s Grok chatbot correctly debunked both the six-fingers claim and a fabricated screenshot alleging Netanyahu’s official account had posted and deleted a death announcement.

Cyabra, a social media threat intelligence firm, reported the broader campaign generated over 145 million views in under two weeks, with TikTok accounting for 72% of engagement. The firm identified tens of thousands of inauthentic accounts with synchronised posting patterns. A separate investigation found 62 fake accounts linked to the IRGC posing as Scottish independence supporters, Irish nationalists, and Latina women. X announced a 90-day revenue-sharing suspension for creators posting undisclosed AI-generated war content, though researchers at the Institute for Strategic Dialogue reported that feeds remained “flooded with AI-generated content about the war” and that the policy had not visibly dissuaded creators.

The campaign sits within a broader pattern. AI-generated imagery of the conflict — fabricated footage of American soldiers captured by Iran, Israeli cities in ruins, US embassies ablaze — has been described by analysts as dwarfing anything seen in previous conflicts. More than 90% of X’s Community Notes are never published, according to the Digital Democracy Institute of the Americas.

The Angle

The six-fingers theory is the detail worth pausing on — not because it was the most consequential part of the campaign, but because of what it reveals about the relationship between detection tools and the audiences they are supposed to serve. AI-generated content has a set of known tells: distorted fingers, inconsistent lighting, anatomical anomalies. These tells have been so widely publicised that they have become the folk epistemology of the internet — the thing people check before they check whether the claim makes sense. The result, in this case, was a real video of a real person being rejected as fake because a shadow on a palm resembled a glitch that people had been trained to look for. The detection heuristic overrode the evidence it was supposed to evaluate.

This is a pattern that scales well beyond one conspiracy about one prime minister. The more fluent a population becomes in the language of AI detection, the more that fluency itself becomes a vulnerability. Once “six fingers means AI” is installed as a reflex, every ambiguous frame becomes available for reinterpretation — and the person doing the reinterpreting feels not credulous but rigorous. They are applying what they were taught. The problem is that what they were taught was a shortcut, and shortcuts survive the conditions that made them useful.

Iran’s operation was effective not because the deepfakes were good — the rubble image had visible compositing artefacts, the fake news presenter couldn’t sync its lips — but because the information environment was primed for them. A conflict that had just produced the assassination of a Supreme Leader, a closing of the Strait of Hormuz, and a prime minister operating under security restrictions that kept him off camera. Into that vacuum, the campaign dropped fabricated images that were bad enough to be detectable and real enough to be shareable, and then let the audience’s own pattern-matching do the distribution work. The 145 million views were not generated by convincing anyone. They were generated by giving people who already suspected something a piece of evidence shaped like confirmation.

The platform response is instructive in what it measures. X’s 90-day revenue suspension targets creators who fail to disclose AI generation. The policy assumes the problem is undisclosed synthetic content. But the most viral element of this campaign was a real video, correctly identified as real by every detection tool that examined it, rejected as fake by an audience that had learned just enough about AI tells to misapply that knowledge at speed. No disclosure policy covers that. No watermark addresses it. The gap is not between real and fake content. It is between the speed at which people learn to recognise fakes and the speed at which they learn when not to apply that recognition.

Netanyahu, for his part, solved it by walking into a café and counting to ten on camera. That it worked is the quietly damning part — not for him, but for the information architecture that required it.