The Reaction Was the Product - Decipon Insights

The Deputy Almost Filed a Report on a Crime That Didn’t Happen

On March 24, 2026, a 22-year-old approached a Seminole County sheriff’s deputy outside an Academy Sports store in Lake Mary, Florida, and showed him a three-second video on his phone. It appeared to show two men entering the deputy’s patrol vehicle in the parking lot. The deputy checked his car. Nobody had touched it. The video was fabricated, generated with AI, and built to look exactly like the kind of clip a bystander hands over as evidence of a crime in progress.

The deputy caught it before it went anywhere. Three days later, a warrant went out for Alexis Martínez-Arizala on four counts: tampering with or fabricating physical evidence, unlawful use of a two-way communication device, false report of a crime, and knowingly giving false information to law enforcement. Sheriff Dennis Lemma’s statement afterward named the actual concern precisely: “The misuse of artificial intelligence to create deepfake videos is a growing concern, particularly when it targets public safety professionals.” He was arrested in Puerto Rico about two weeks later.

He kept making the videos.

A Timeline of Manufactured Emergencies

By October, Martínez-Arizala was back in Florida and back at a Home Depot in Palm Beach Lakes, showing a shopper named Melanie Valentine a fabricated video of her husband’s truck being stolen from the parking lot — footage realistic enough that she described it as “very real.” He urged her to follow him outside to “catch the guy.” She didn’t. She later said what stuck with her wasn’t the video’s quality, it was the ask: “I had a young guy run up to me, shoving his phone in my face.”

On November 18, he approached a different shopper’s wife, showed her an AI-generated video depicting her husband kissing another woman, and recorded her reaction on a pair of Meta glasses. She told a reporter afterward: “I just felt violated — that was the worst part.” Police responded to the scene. His explanation was two words: “It’s an AI prank.”

West Palm Beach police say they know of at least three incidents matching this pattern at the same store. On November 30, the same man approached a clerk at a Marathon gas station with another fabricated video and was arrested — not for the video, but for resisting the officer who tried to detain him. Bodycam audio picked up his read on the whole night: “I just made a banger with this one, boss. My last one on TikTok hit, like, a mil.”

Two arrests, eight months, three named victims, an unnamed number of others police are still trying to identify. Not one of them a public figure, an election, or a geopolitical flashpoint. Just a phone, a generator, and whoever happened to be standing in a Home Depot parking lot.

Why These Targets, Specifically

None of the videos were random in subject. A spouse’s infidelity, a stolen vehicle, an officer’s own patrol car — each one was selected for the single emotion that produces the most watchable reaction on camera. Infidelity produces grief and rage in real time, aimed at a real marriage. Theft produces adrenaline and a plausible reason to comply with a stranger’s instructions — get in the car, come with me, hurry. An officer’s stolen cruiser produces the sharpest irony available: fool the person whose job is not to be fooled.

This is the same targeting logic that shows up in every fabricated-evidence operation Decipon has covered, just run at the scale of one phone instead of one government. The Influence Tactics Protocol treats manufactured “proof” — a doctored photo, a laundered quote, a synthetic video — as a category because the mechanism doesn’t change with the target’s size: give a person something that looks like visual evidence, aimed at their strongest attachment (a spouse, their property, their competence), and the reaction is authentic even though the premise is not. Martínez-Arizala didn’t need a narrative or an ideology. He needed a wife who loved her husband and a woman who owned a truck.

“It’s an AI Prank”

The minimizing label did work that’s worth naming precisely, because it’s the same rhetorical move disinformation operators reach for when caught: reclassify an intentional deception, after the fact, as a genre that isn’t supposed to be taken seriously. “It’s satire.” “It’s just a meme.” “It’s an AI prank.” The frame doesn’t undo the manipulation — the woman who was shown her husband kissing someone else still felt violated after the reveal — it just tries to make the manipulator’s exposure feel like an overreaction on the victim’s part rather than a harm on his.

It’s a defense built for the moment of confrontation, not for the video itself. Nobody consented to be manipulated into a strong emotional reaction so it could be uploaded; “prank” is a word applied afterward, to whoever gets upset about that fact.

The Incentive Nobody Had to Hide

What separates this case from most manipulation Decipon covers isn’t the mechanism — it’s that the incentive structure is fully unmasked. State-linked influence operations tend to bury their engagement-chasing behind a claimed cause: national security, a political outcome, an ideology worth spreading. Martínez-Arizala said the quiet part into a body camera. “In one month I hit 50,000 followers on TikTok.” “My last one on TikTok hit, like, a mil.” Police reports note he told them he believed people would appreciate the content more “because he’s willing to get arrested.”

Strip away the ideological packaging that usually surrounds fabricated-evidence campaigns and this is what’s underneath: a platform that pays out in attention for whatever provokes the strongest visible reaction, and a person optimizing directly for that payout with no cover story required. The Home Depot videos and a state-run fabrication campaign are drawing from the same well — content that performs well is content that manufactures a strong emotion, verified or not — the only difference is that one operator had a flag to hide behind and the other had a follower count to chase.

When Looking Harder Doesn’t Work

Greg Gogolin, who directs Ferris State University’s Center for Cybersecurity and Data Science, made the uncomfortable part of this explicit: modern video generators need barely any source material — “a snapshot” can be enough — and the old detection tells, unnatural blinking, stiff mannerisms, mostly don’t hold up anymore. Lip-sync mismatches and off facial expressions might still give something away, but that’s a narrowing target, not a growing one.

Which means the tell in these incidents was never in the pixels. Valentine didn’t catch the fabrication by staring harder at the footage — the video, by her own account, looked real. What she caught was the request riding on top of it: a stranger she’d never met, asking her to leave the store and get closer to him because of what she’d just seen. That’s the part a generator can’t fabricate away, because it isn’t in the video. It’s in the ask.

What This Is Not

Worth being precise about the limits here. Martínez-Arizala has not been convicted on the Seminole County charges — a warrant and an arrest are not a verdict, and he’s entitled to the presumption that comes with that. Florida police have also been explicit that the Home Depot incidents themselves haven’t produced AI-specific charges; his November 30 arrest was for resisting an officer, not for the video that led to the encounter. This isn’t a story about a person already held fully accountable by the legal system for the manipulation itself — it’s a story about a person who kept fabricating videos of strangers for months after his first arrest for doing exactly that, and about how thin the deterrent turned out to be.

It’s also not a story about political disinformation, and forcing it into that frame would overclaim. There’s no ideology here, no state sponsor, no election. That’s precisely why it belongs in the same analysis as the operations that do have those things: it shows the underlying mechanism with nothing else in the way.

The Pattern Worth Watching

Manufactured evidence plus targeted emotional exploitation plus a reward structure that pays for reaction over truth — that triad is the whole operation, whether it’s run by an intelligence service with a narrative to advance or a 22-year-old with a phone and 50,000 followers. Remove the ideology and the mechanism doesn’t weaken. If anything it gets cleaner to see, because there’s no cause left to argue about, just the transaction: someone else’s genuine fear or heartbreak, converted into content, for an audience that will never know the difference and mostly won’t ask.

The defense that held up in these incidents wasn’t better video literacy. It was Valentine noticing that the request mattered more than the footage. As synthetic video gets harder to catch by looking at it, that’s the shift media literacy has to make with it — from “does this look real” to “what is this person, right now, asking me to do.”

This article is part of Decipon’s Manipulation Breakdowns series, examining specific influence operations through the Influence Tactics Protocol.

Sources: