AI-Generated Video Falsely Depicting Large UK Pro-Palestine March Spreads on Social Media

A video depicting what appeared to be an enormous pro-Palestine march filling the streets of the United Kingdom circulated on social media in 2026, drawing widespread engagement before independent fact-checkers established that the footage was entirely fabricated. Two organisations — Full Fact in the United Kingdom and Thai PBS Verify in Thailand — independently confirmed that the video was not documentary footage of any real event but rather an AI-generated fake video containing detectable digital watermarks left by Google’s AI tools. The case has become a pointed illustration of how synthetic media can exploit genuine political tensions to spread false narratives across borders, and of why technical watermarking standards matter in the current information environment.

What the Video Showed and How It Spread

The video presents a striking visual: an immense crowd moving through what is implied to be a British city street, framed beneath two enormous flag banners — one Union Jack and one Palestinian flag. The imagery is designed to convey scale and solidarity, suggesting a mass demonstration of public sentiment. A caption accompanying one Facebook post read “UK peoples Stand with Palestine 🇵🇸 free Palestine free Gaza”, directly inviting viewers to interpret the footage as evidence of widespread British public support for the Palestinian cause.

On the platform Threads, a user identified as ‘torontopost11’ posted a version of the video on 21 May 2026 with the caption “London UK Rises for Palestine — Massive Rally Fills the Streets”, according to Thai PBS Verify’s analysis of that post. The framing across platforms was consistent: the video was presented as authentic documentation of a real, large-scale street demonstration in the UK. That framing is precisely what made it effective as a piece of disinformation — it attached itself to a genuine and ongoing political controversy, lending the fabricated imagery a veneer of plausibility.

It is important to note that actual pro-Palestine rallies have taken place in the United Kingdom. Thai PBS Verify confirmed that real demonstrations did occur, which means the video’s claim was not wholly disconnected from reality — it was, rather, a fabrication built on top of a real political context. That layering of false content onto true background conditions is one of the reasons AI-generated disinformation can be particularly difficult for ordinary viewers to dismiss on instinct alone.

The Technical Fingerprints: Google’s SynthID Watermark

The clearest and most technically significant finding from the fact-checking investigations was the presence of Google’s SynthID watermark in both the audio and visual components of the video. Full Fact’s analysis, published on 29 September 2026, confirmed this detection. SynthID is a watermarking system developed by Google DeepMind that embeds imperceptible signals into AI-generated content — signals that survive compression, re-encoding and the kind of processing that typically accompanies social media sharing. The presence of SynthID in both the audio track and the visual frames indicated that the content was edited or generated using Google AI tools.

This is a meaningful technical detail. SynthID does not appear in content that was simply filmed on a phone or edited in conventional video software. Its presence is a direct indicator that AI generation or AI-assisted editing was involved at the production stage. For fact-checkers and platform trust-and-safety teams, this kind of embedded signal provides a more reliable basis for classification than purely visual analysis, which can be subjective and contested.

Full Fact’s use of SynthID detection as part of its verification process represents an important methodological development. Rather than relying solely on contextual clues or reverse image searches — techniques that remain valuable but can be circumvented — the organisation was able to point to a specific, technical, source-level indicator of AI generation. Full Fact’s published analysis of the video provides the detailed findings for readers who wish to examine the evidence directly.

👉 Read also: AI-Generated and Manipulated Videos Spread False Claims About UK Protests, Asylum Policy, and Disasters

Visual Glitches as a Secondary Indicator

Beyond the SynthID watermark, the video itself contained observable artefacts consistent with AI-generated imagery. Full Fact noted that the footage includes notable glitches, with unnatural movement throughout the crowd and some smaller flags appearing distorted. These visual anomalies are characteristic of current AI video generation tools, which can produce convincing large-scale compositions but frequently struggle with fine-grained details — particularly objects in motion, text, and repeated small elements like flags in a crowd.

For a trained observer, these glitches can serve as red flags. Flags, in particular, are a known weakness of many AI video generation systems: their movement is complex, involving fabric physics, wind interaction and perspective shifts, and current models often render them in ways that look subtly wrong — edges that blur inconsistently, fabric that folds in geometrically implausible ways, or motion that lacks the randomness of real cloth. The distorted smaller flags noted in this video fit that pattern precisely.

However, it would be a mistake to rely on visual glitches alone as a detection method. AI video generation is improving rapidly, and artefacts that are obvious in 2026 may be invisible in subsequent model generations. The SynthID watermark, embedded at the point of generation, is a more durable indicator — which is why the development and adoption of such watermarking systems is increasingly important to the broader ecosystem of media verification.

Geographic Dislocation and Cross-Border Disinformation

One of the more striking aspects of this case is the geographic spread of the video’s distribution. Thai PBS Verify — a fact-checking unit based in Thailand — identified and analysed a version of the video posted by a Threads user whose handle referenced Toronto, Canada. The video itself purported to show events in the United Kingdom. Full Fact, a UK-based organisation, published its own independent analysis. The claim, in other words, was not geographically contained: it circulated across multiple national contexts, picked up by users in different countries who may have had no particular familiarity with what a real UK protest looks like or how it would typically be documented.

This cross-border dimension is not incidental. Thai PBS Verify noted that the video circulated internationally, with the Threads post by ‘torontopost11’ being one documented instance of that spread. When fabricated content about events in one country circulates primarily among audiences in other countries, those audiences have less access to the contextual knowledge that might help them identify inconsistencies. They cannot, for instance, recognise that a particular street does not match any known London thoroughfare, or that the crowd density depicted would be extraordinary even by the standards of the largest documented UK demonstrations.

Thai PBS Verify confirmed that while genuine pro-Palestine demonstrations did take place in the UK, the specific video and the overwhelming crowd size it depicted were fabricated using AI. That distinction — between the existence of real events and the fabrication of specific footage — is central to understanding how this type of disinformation operates. It does not invent a controversy from nothing; it manufactures false evidence for a controversy that already exists, making the fabrication harder to reject and easier to share in good faith.

Why AI-Generated Fake Videos Target Political Demonstrations

The choice of a pro-Palestine march as the subject of this AI-generated fake video is not random. Footage of mass demonstrations carries particular rhetorical weight in political discourse: crowd size is routinely used as a proxy for the strength of public feeling, and images of large gatherings are among the most shared types of political content on social media. An AI-generated video that appears to show hundreds of thousands of people marching in solidarity makes an implicit argument about the scale and legitimacy of a political position — an argument that would be considerably harder to make through text or statistics alone.

👉 Read also: AI-Generated and Mislabeled Videos Spread False Claims About Protests, Arrests, and Public Events

Demonstrations are also events that people expect to be documented on video. The existence of smartphone footage of protests is so normalised that a video claiming to show one carries an immediate presumption of authenticity. Unlike, say, a fabricated video of a political leader giving a speech — where viewers might wonder why they have not seen the clip in mainstream news coverage — a video of a street march can plausibly be explained as amateur footage that “went viral” without passing through editorial filters. That plausibility is a structural vulnerability that AI-generated content can exploit.

The political context matters too. The Israeli-Palestinian conflict is among the most intensely debated international issues in global public discourse. Content related to it generates high engagement, strong emotional responses and rapid sharing across ideological lines. A fabricated video that appears to document significant public support for one side of that debate is likely to be shared enthusiastically by those who agree with its implied message and angrily by those who do not — either way, the fabrication achieves broad circulation.

The Role of Watermarking in Verification

The detection of Google’s SynthID watermark in this case points to a broader conversation about the role of technical standards in combating AI-generated disinformation. SynthID was developed specifically to address the problem of AI-generated content circulating without any indication of its synthetic origin. By embedding an imperceptible but detectable signal at the point of generation, the system creates a traceable record that survives the re-encoding and compression that content typically undergoes as it moves across platforms.

Full Fact’s ability to confirm the SynthID watermark in both the audio and visual components of this video demonstrates that the system works in practice, not just in laboratory conditions. That is a significant finding. It suggests that watermarking can serve as a reliable technical layer in the verification process — not a replacement for editorial judgement, but a complement to it.

At the same time, watermark detection is only useful if fact-checkers and platform moderators have the tools and training to apply it. The two organisations that identified the SynthID signal in this video — Full Fact and Thai PBS Verify — are specialised fact-checking units with technical capacity that most social media users and many newsrooms do not possess. Scaling watermark detection to the volume of content that circulates daily on major platforms remains an unsolved problem, and the presence of a watermark does nothing to slow the spread of content before it is detected.

What Fact-Checkers Found and When

Thai PBS Verify identified a version of the video — specifically the post by Threads user ‘torontopost11’ dated 21 May 2026 — as part of its investigation. Full Fact published its own independent analysis on 29 September 2026. The two investigations reached consistent conclusions through independent means, with both confirming the AI-generated nature of the footage. The fact that two organisations operating in different countries and media ecosystems arrived at the same finding through separate processes strengthens the reliability of the verdict.

Neither organisation was able to establish the identity of the original creator of the video, and the exact date on which the video first appeared online remains unknown. These gaps are typical in cases involving AI-generated content: the tools used to create such material are widely available, accounts that share it are often newly created or pseudonymous, and the speed of social media distribution means that content can accumulate significant reach before any systematic investigation begins.

The case of this fabricated pro-Palestine march video is, in the end, a clear instance of a pattern that is becoming more common and more technically sophisticated: AI-generated fake video content attached to real political events, distributed across national boundaries, and designed to be visually compelling enough to circulate before verification can catch up. The detection of Google’s SynthID watermark in both the audio and visual components gave fact-checkers a definitive technical basis for their verdict, and the convergence of Full Fact’s and Thai PBS Verify’s independent analyses leaves no credible doubt about the footage’s synthetic origin. What the case also demonstrates is that the infrastructure for detecting such content — watermarking systems, trained fact-checkers, cross-border collaboration — is developing in parallel with the tools that produce it, and that the gap between production and detection, while still real, is narrowing.

This article was produced with AI assistance and reviewed editorially.