Manipulated and recycled photos and videos are common during wars and crises — sometimes intended to deceive, sometimes meant only to attract clicks. During the US‑Israel war with Iran this problem escalated when manipulated and apparently AI‑generated images supplied to international photo agencies spread into European newsrooms. Some fakes were produced by artificial intelligence, others were digitally altered by people. The result: hundreds of suspect images circulated widely before outlets discovered the fraud and pulled them down.
The SalamPix episode
In early March, Dutch news agency ANP removed roughly 1,000 Iran‑related photographs from its database after flagging them as potentially AI‑manipulated. Two days later, RTL Nieuws acknowledged that it had unintentionally used three of those images on its website and app; after being alerted by ANP, RTL removed the photos and published a notice explaining which images had been taken down and why. German weekly Der Spiegel also admitted it had published an image later judged to be AI‑manipulated.
The images reaching ANP and other providers traced back through a chain of agencies: Abaca Press supplied material to international distributors such as dpa Picture Alliance, ddp and Imago Images, which in turn distributed pictures to outlets including ANP and others. The source feeding Abaca was identified as an Iranian agency, SalamPix. In response, several photo agencies blocked SalamPix content or issued “kill notices” instructing clients to remove those images from publication.
Why reputable agencies were fooled
Many news organizations rely on the longstanding practice — and, in some jurisdictions, legal protections — of trusting material provided by verified news agencies. In Germany, for example, an “agency privilege” lets outlets rely on the authenticity of verified text, images and videos from agencies. Global broadcasters and newsrooms routinely depend on these external services to cover events worldwide.
That model is increasingly strained by the rise of sophisticated AI tools. As fabricated images grow more convincing, spotting fakes becomes harder, especially during fast‑moving breaking news when editorial teams must process massive volumes of visuals quickly. Deutsche Welle reports that since the start of 2026 it has received about 140,000 images per day from agencies — a volume that complicates thorough manual vetting.
Mathias Stamm, DW’s editor‑in‑chief, emphasizes transparency: DW aims to make any AI‑generated content clearly identifiable, and when mistakes happen — as with SalamPix images — the outlet acknowledges them and corrects the record.
What DW and others found
After the SalamPix revelations, DW reviewed its coverage, removed SalamPix images, and published corrections where necessary. Analysis turned up several telltale signs of AI or digital manipulation. One image presented as a Tehran street scene after a missile strike showed apparently realistic cars, buildings and smoke, but closer inspection revealed clear errors: wall and vehicle signage that looked like writing at a distance but resolved to nonsensical pseudo‑text, bulging or warped walls and windows that should be straight, and vehicles with odd, impossible shapes.
Another image, captioned to show security forces firing on protesters, depicted a man in black holding a weapon. Zooming in exposed inconsistencies: the two shoes differed in size and shape, the shadow and limb positions didn’t match the body, and a hand showed an anatomically incorrect gap between thumb and fingers. Other SalamPix images contained deformed fingers, misaligned windows and distorted faces — classic artifacts associated with early and mid‑generation AI image models and some manual edits.
How newsrooms and the public can spot fakes
As AI tools improve, both journalists and the public need new habits and tools. Measures that newsrooms are using or recommending include:
– Training journalists and photo editors to recognize common AI artifacts — inconsistent shadows, warped geometry, nonsensical text, duplicated or mutated body parts.
– Instituting verification workflows for agency content, including reverse image searches, metadata checks and inquiries to source agencies about provenance.
– Flagging and clearly labeling any content known to be AI‑generated, and issuing prompt corrections when manipulated images are published by mistake.
– Producing media literacy resources so audiences can spot likely fakes themselves and understand corrections when they appear.
DW and other media organizations are investing in staff training and fact‑checking teams for exactly these reasons, and some outlets are tightening rules about accepting third‑party visuals.
The SalamPix incident underscores how quickly manipulated imagery can move through established distribution chains and into trusted newsrooms. It also highlights the need for more rigorous verification, clearer labeling of AI content, and ongoing media literacy efforts to help both professionals and the public navigate an environment where realistic fakes are becoming commonplace.
Edited by Thomas Sparrow, Joscha Weber and Cristina Burack