Quick answer: Belgian agency AiCandy posted a hyper-realistic AI deepfake ad in March 2026 showing Elon Musk, Jeff Bezos, and Sam Altman pitching a fictional Energym startup that harvests energy from laid-off humans pedaling bikes to power AI. The creator confirmed it was satire. It went viral on X, and many viewers did not get the joke.
In early March 2026, a 90-second video began circulating on X showing aged-up, hyper-realistic versions of Tesla CEO Elon Musk, Amazon founder Jeff Bezos, and OpenAI CEO Sam Altman pitching a new venture called Energym. The premise: a Matrix-style gym chain where laid-off workers pedal stationary bikes and row machines to generate the electricity that powers the AI systems that put them out of work. The footage was documentary-style, with all three "executives" giving talking-head interviews about their startup.
The video was created by AiCandy, a Belgium-based creative video agency founded in 2025 by Jan De Loore (42) and Hans Buyse (52). It was satire. De Loore confirmed the framing in Fortune's coverage: the three executives were chosen because "they are the face" of AI, and the goal was to embody the job-loss anxiety he believed people were carrying about it, layered on AI's well-documented energy footprint. The video carried no on-screen disclosure that it was AI-generated.
A meaningful share of viewers thought it was real.
This post walks through why the AiCandy video is the canonical example of the share-without-verifying problem on X specifically, what makes X structurally different from Instagram or TikTok for AI-content diffusion, and what the satire-vs-disinformation overlap means for what you should actually do when something high-engagement crosses your feed.
For the detection-skill side of the same problem, see the seven-tells guide for X posts. For the broader technical grounding, see the pillar guide on what a deepfake actually is.
A two-person Belgian agency, three billionaires, one viral satire
AiCandy is run by founders Jan De Loore and Hans Buyse as a two-person creative shop. The Energym deepfake went viral on X after posting in early March 2026, with viewers in the replies asking whether the startup was real and others sharing it as confirmation of the AI-driven future they already feared.
Source: Fortune coverage of AiCandy's Energym deepfake, March 2026.
Why X Is the Hardest Platform for AI-Content Discernment
Three structural reasons the AiCandy video spread the way it did on X specifically.
The "Made with AI" label depends on the creator opting in. X's AI labeling regime, rolled out across 2026, requires the creator to toggle the label on at post time, or for C2PA Content Credentials to be present in the file. AiCandy did not toggle the label. C2PA was stripped (or never embedded) at upload. The post carried no platform-level signal that it was synthetic. The same gap that defeats Instagram's AI Info label and Meta's broader labels-over-removal posture applies here, but with a twist: on X, creators of obvious satire are even less likely to apply the label than creators of straight disinformation, because applying it breaks the satirical conceit.
Community Notes lag virality by hours to days. Community Notes is X's distinctive moderation layer, and the slowest-acting one. Notes require multiple raters from diverse perspectives to agree on a note before it appears publicly. For a fast-moving video, the note often arrives after the post has already been shared into a viewer's feed dozens of times. The first 24-48 hours of a viral post on X are functionally unlabeled, even when a note is eventually attached.
The retweet-quote-stack rewards confident framing over verified framing. A post's reach on X scales with the engagement it gets in the first hour, and engagement scales with the confidence and stake of the quote-retweets layered on top. A skeptical quote-tweet ("is this real?") generates less engagement than a confident one ("THIS is the dystopia they want"). The algorithm's reward structure systematically prefers stake-taking over verifying. The AiCandy video benefited from this exactly: confident quote-retweets treating the startup as real outperformed skeptical ones asking whether it was a parody.
For the analysis of how political and election-cycle deepfakes hit the same diffusion pattern, see the 2026 midterms AI political ad post and the Iran airman case study.
Why Satire Does Not Solve the Diffusion Problem
The first defense people reach for is "but it's satire, so it's fine." The AiCandy case is the cleanest counterexample available in 2026.
Satire requires both parties to be in on the joke. When the satirical framing depends on visual hyper-realism (as the AiCandy video did, deliberately, because the executives looking real was the point), the satirical signal lives at the metadata layer, not the perceptual layer. Strip the metadata (which X's algorithm and the share-flow do for free) and the viewer is left with a hyper-realistic video of three named public figures making specific claims about a startup that does not exist. The viewer cannot perceive the satire if the satire was visual. The viewer can only infer it from context that the share-flow stripped away.
De Loore's Fortune interview makes the underlying anxiety explicit. The satire worked because the anxiety was real. That is what made the framing effective. It is also what made the diffusion irreversible: the same viewers most worried about AI-driven job loss were the viewers most primed to share the video without checking whether the startup was real, because the startup matched the future they already feared.
This is the same diffusion pattern that drove the Iran airman AI image to be shared by three sitting US politicians. The image fit the worldview; the worldview defeated the verification.
What This Means for You on X
Three rules that apply when high-engagement content crosses your feed.
Treat the absence of a "Made with AI" label as zero information. X's labeling regime is creator-opt-in plus C2PA-dependent. Real AI content routinely shows up unlabeled. A label being present is a positive signal that the content is AI-generated; a label being absent tells you nothing. Do not let "no label" function as "verified real."
Search for the entity in a non-X source within 30 seconds. For the AiCandy video, "Energym Musk Bezos Altman" returns either the Fortune coverage of the deepfake or nothing. Either result tells you the same thing: this is not a real company. The X-internal context (replies, quote-retweets, Community Notes) is the slowest signal. Going outside X to a search engine or a reputable outlet is the fastest. Thirty seconds, one search, before you share.
Apply the seven detection signs from the X-post guide. Most AI video on X is not produced by a Belgian creative agency with weekend-grade satire goals. Most is produced by scam accounts, political operators, or content farms with worse craft and clearer monetization motives. The seven-tells guide for X posts covers the visual and account-level signals that catch the typical case. The AiCandy video is the high-craft outlier; the rest is the bulk.
If a viral post on X caught your attention recently and you want to check whether the account behind it has been flagged by the Ledger community, paste the URL below.
Think you found an AI video?
Paste the URL and let the Ledger community verify it. Free.
The Bigger Pattern
The AiCandy case sits at the intersection of three trends that will keep producing similar incidents. Generative video has crossed the hyper-realism threshold and is now within reach of small creative shops. Platform labels are voluntary and lag content. The X engagement loop rewards confident framing over verified framing. None of these trends are reversing on a timescale that matters to a viewer scrolling tomorrow.
What changes is the consumer-side defense layer. The detection-skill guides on this site exist because the platform-side defense has structurally failed. The label is a positive signal when present and zero information when absent. Community Notes catch some cases hours after the damage. Outside-X verification (a 30-second search) remains the fastest reliable check.
For the X-platform-specific detection workflow, the seven tells guide is the load-bearing reference. For the broader pre-share protocol, the 5-minute verification guide covers the workflow for any platform. AiCandy gave the case study; the workflow is what stays once the news cycle moves on.
Related Posts
- How to Tell If an X (Twitter) Post Is AI-Generated: 7 Signs to Check Before You Share: the X-specific detection sibling that covers the visual and account-level signals the AiCandy case spotlights
- What Is a Deepfake? A Plain-English Guide for Social Media Users: the technical foundation explaining how generative video reached the hyper-realism threshold the AiCandy case operates at
- Three US Politicians Shared an AI Image as Real: The Iran Airman Incident: the diffusion-pattern sibling where worldview-confirming AI content defeated the verify-before-sharing principle in elected officials' hands
- How to Tell If a TikTok Video Is AI-Generated: The Complete Guide: the cross-platform detection-skill pillar that the AiCandy case validates as load-bearing wherever AI labels lag synthesis
