Seeing things

• The scale. The social-media-intelligence company Cyabra traced identical videos, synchronized posting windows, and fixed hashtag clusters to centralized Iranian content production across TikTok, X, Facebook, and Instagram. TikTok alone accounted for 72 percent of total views and more than seven million engagements. The content arrived simultaneously in Persian, Arabic, Urdu, and English. This wasn’t improvised. It effectively launched alongside the missiles.
• The satellite front. A new category of fabrication: AI-generated satellite imagery designed to mimic the visual authority of open-source intelligence. Iran’s state-aligned Tehran Times posted a “before and after” image claiming to show a destroyed American radar facility in Qatar. Researchers found it was a doctored Google Earth image of a U.S. base in Bahrain, taken in February 2025—same cars, same parking spots, a year apart. AFP detected Google’s invisible SynthID watermark, confirming AI manipulation. Imposter OSINT—open-source intelligence—accounts have appeared on X, mimicking the format of legitimate digital investigators. The verification method that arose to circumvent government censorship during conflicts is now a target.
• Not your grandad’s propaganda. The old wartime information model—two sides, each broadcasting a version of events, from Lord Haw-Haw to the Gulf War—assumed a shared visual record that competing sides interpreted differently. The Iran war doesn’t have one. It’s a simultaneous, multilateral information environment in which Iranian state campaigns, White House meme videos, Israeli AI-processed intelligence, Russian targeting data, commercial engagement farmers, and automated bot networks all occupy the same feed. Cyabra caught Iranian bot networks that had previously posed as British users stoking Scottish separatism—1,300 fake profiles, 224 million potential views—pivoting overnight to pro-Iranian war narratives when Iran’s internet came back on. A NATO Strategic Communications Centre of Excellence report, published weeks before the war, concluded that AI-enabled influence operations are now faster, more persuasive, and harder to detect than anything that preceded them.
• The engagement economy. Not all of this is statecraft. BBC Verify, the broadcaster’s fact-checking unit, found that a significant share of AI-generated war content comes from creators farming views and revenue—cutting fabricated clips to trending audio, captioning them in multiple languages, designed to perform rather than persuade. State propaganda and commercial grift now use the same tools, the same platforms, and the same incentive structure. The conflict’s visual record is, in no small part, a commercial product.
• The closed loop. When users on X turned to Elon Musk’s AI chatbot Grok to check whether footage was real, Grok repeatedly confirmed fabricated videos as genuine—citing Newsweek and Reuters to support conclusions those outlets’ reporting didn’t warrant. AI generates the fakes. AI confirms them. The architecture of X—which rewards engagement over accuracy—ensures they circulate. The verification layer is producing its own misinformation.
• The inversion. Factnameh, an independent Persian-language fact-checking outlet run from Toronto that has tracked Iranian-regime disinformation since 2017, observed a pattern accelerating after last June’s 12-Day War—the U.S.-Israeli strikes on Iran: As AI fakes multiply, people dismiss real evidence that contradicts their expectations. They reject authentic footage; they believe the fabrications. The volume of fakes has produced a permission structure for disbelieving anything. The technology doesn’t just manufacture false images. It’s eroding the authority of genuine ones.
• The accountability gap. Congress voted not to require the administration to seek authorization for continued military operations—less a debate about the war’s purpose than a test of party loyalty. The administration’s stated objectives have expanded from four to at least five without anyone in a position of authority requiring a definition of success. The visual record on which any public reckoning would depend—what was hit, what was destroyed, who was killed, what the war looked like—is itself unreliable. And on Saturday, Brendan Carr, the chair of the U.S. Federal Communications Commission—the agency that regulates American broadcasters—threatened to cancel the broadcast spectrum licenses of news outlets he accused of “hoaxes and news distortions” about the war. U.S. Defense Secretary Pete Hegseth suggested CNN’s coverage would improve once David Ellison—who reportedly assured the White House that his overhaul of the network would result in its coverage being more favorable toward the administration—completed his acquisition of its parent company. While hundreds of AI-generated fakes circulate unchallenged, the administration’s regulatory pressure appears to be aimed at the reporters trying to sort real from fabricated.

Every prior American conflict had a visual record that anchored public understanding—however contested, however partial. Saigon had its photographs. The Gulf War had its night-vision video. Iraq had footage of Abu Ghraib. Those images could be argued about, but they existed as shared points of reference: This happened, and here’s what it looked like.

The Iran war, two weeks in, is producing something different—and abjectly disorienting. Its most widely seen images are fabrications. Its official communications treat airstrikes as memes. People dismiss its authentic footage as inauthentic. And the AI tools that generate the fakes are improving faster than the tools built to detect them—BBC Verify’s Sardarizadeh, one of the most prominent conflict-disinformation analysts working, says new fakes appear faster than anyone can flag them.

Millions of us are watching this war. A lot of what we’re seeing—produced by governments, bot networks, and engagement farmers, often indistinguishably—isn’t real. Since the invention of the camera, at least, people have had to try to understand conflicts based on incomplete and sometimes misleading visual evidence. Now the evidence is abundant, vivid, and arrives in hours—and the false material looks more like what many of us expect reality to look like than the real footage does. The immediate struggle, in this informational environment—the practical struggle we’re faced with—isn’t to understand what’s happened, even though that’s what democratic accountability requires of us. It’s to figure out how many of the 50 things we just watched didn’t happen.