When the Feed Is Fake: AI-Generated Video and the Food System We Trust

We already know what a viral undercover video does to a farm operation. Now imagine the footage never had to be filmed.

I spent last year writing a book about cybersecurity in food and agriculture. Ransomware hitting meat processors, IIoT concerns in fisheries, phishing attacks on co-ops, GPS spoofing on precision ag equipment, OT vulnerabilities in food factories. A f’ton to cover, and I’m proud of it! What didn’t make it into those pages is what I want to talk about here, not because it isn’t real, but because the landscape was moving faster than I could pin it down with sourcing I trust. The threat sits somewhere between “emerging” and “imminent,” and I have a personal rule not to write things I can’t back up, or that are evolving without much coverage. Well, now I can.

The Proof of Concept Already Happened

In late October 2025, a wave of AI-generated videos started circulating on TikTok and Facebook. They depicted Black women, synthetically generated, ranting about SNAP benefit cuts during the government shutdown debate. One character claimed to have “seven different baby daddies.” Another threatened grocery stores. The original posters labeled the content as AI-generated right in the post, making it visible to everyone.

Fox News Digital ran a story treating them as real anyway. The headline read: “SNAP beneficiaries threaten to ransack stores over government shutdown.” After being called out by multiple journalists, the outlet quietly rewrote the piece at the same URL, with the same timestamp, and swapped the headline to acknowledge that the videos were AI-generated. No formal retraction, only a small editor’s note. Meanwhile, a broader wave of similar videos accumulated millions of views during that period, content engineered to inflame opinion around a program that feeds approximately 42 million Americans.

And this was not a targeted attack on a farm or a food brand. Nobody’s supply chain went down. No processing line stopped. And yet it reshaped public perception of who benefits from food assistance, who “deserves” food, and what behavior looks like in a grocery store. That’s a food system attack. It just arrived wearing clothes different from the ones we’re trained to recognize.

What Deepfakes Are Actually Doing to This System

I think about this problem from a systems perspective: Food and agriculture are cyber-physical systems. The physical world, fields, barns, aquaculture, processing lines, cold chains, and distribution networks are tightly coupled with digital infrastructure. What happens in the digital space has physical consequences, and the reverse is equally true.

Deepfakes don’t need to compromise a SCADA system or get anywhere near an Industrial System to cause real damage. They go after what I’d call the perception layer, the part of any system that decides what is true, what happened, and what should happen next. A bank can absorb a rumor the way a thick-walled silo absorbs weather from a storm. A farm subjected to viral videos of alleged abuse of their animals is in a different situation entirely. The damage travels before the truth can get its boots on.

Think about what a convincing fake video could set in motion here. A fabricated “undercover” animal abuse clip at a named livestock facility doesn’t need to be technically flawless. It needs to be believable enough to spread during the few hours before anyone with authority can debunk it. We already have a very clear picture of what real undercover footage does. In 2019, Animal Recovery Mission released hidden-camera footage showing systematic calf abuse at Fair Oaks Farms in Indiana. Within days, Jewel-Osco, Strack & Van Til, Family Express, and Tony’s Fresh Market had pulled Fairlife products from their grocery shelves. Three former employees faced criminal charges. Civil litigation settled for $21 million in 2022. That was footage captured by a real investigator inside a real facility over months of work. The deepfake version skips the months entirely, and you don’t have to leave your house.

A fake food safety video showing contamination in a processing facility doesn’t need a plot. Products on a production line where they shouldn’t be is enough to trigger consumer panic, retailer calls, and regulatory inquiries before anyone at the facility has confirmed the footage is synthetic. The phone rings first. The investigation comes considerably later.

A deepfake of a spokesperson, a CEO, a USDA official, or a well-known farmer can announce a recall that never happened or admit to practices they’ve never engaged in. Voice cloning and video synthesis now cost about as much as a streaming subscription, which was roughly where email phishing was in terms of accessibility around 2004. We know how that story went.

I don’t have to invent nightmare scenarios; reality’s doing great on its own. In 2025, TikTok accounts using OpenAI’s video generation tools fabricated clips of Indian street food vendors in grotesque, unhygienic scenarios. One video hit 40 million views. The vendors depicted were real people running real businesses. The footage was entirely synthetic, and the damage to their reputation was not. The mechanism works. It just hasn’t been pointed squarely at a farm operation yet.

Why Food and Agriculture Is Sitting in Crosshairs

Other industries face deepfake risk. What makes food and agriculture more exposed comes down to a few things worth naming plainly rather than gesturing at vaguely.

Food is emotional in a way that logistics or industrial manufacturing simply isn’t. It touches health, culture, religion, family memory, animal welfare, environmental anxiety, and economic identity all at once; often in the same bite. Think of it like a fire alarm in a crowded building: once it’s pulled, people react immediately, moving, warning others, and assuming the threat is real. The connection people have to their food, to farmers, and to the stories about where things come from means that once that alarm is triggered, the response spreads quickly and is hard to reverse, even if it was a false alarm. Content engineered to trigger disgust or fear about food moves through social networks faster than almost any other category of misinformation, because the gut reacts before the skeptical part of the brain has a chance to object.

Visual evidence also carries disproportionate weight, specifically in this sector. Undercover videos have been devastating for livestock producers, not primarily because of their legal value as evidence, but because humans are cognitively wired to treat video as witness testimony. We’ve outsourced our sense of “being there” to the camera. AI-generated video exploits that wiring, and nothing in the format itself flags its origin. There’s no synthetic smell, no uncanny valley moment if the generation quality is good enough. It just looks like something that happened.

The attack surface runs unusually long, too. A deepfake targeting a single farm can ripple through its seed supplier, processor, retail buyers, certifying body, and the agencies that oversee it, without any of those entities’ systems being touched. The integration that makes the food supply chain efficient is the same property that lets reputational damage travel far and fast. It’s like having excellent plumbing; very useful until something toxic gets into the pipes.

Then there’s the detection problem, which is uncomfortable to explain but important to be honest about. In March 2025, CSIRO tested 16 leading deepfake detection tools against real-world content and found that average accuracy hovered around 55 percent, which is statistically indistinguishable from guessing. A separate benchmark studied the same month found that open-source detection model accuracy dropped roughly 50 percent when tested outside controlled academic conditions. The core issue is what researchers call a generalization gap: tools trained to catch fakes generated by one method fail when the generation method changes. Given that AI video tools are improving roughly every quarter, that gap is structural. Buying a detector and calling the problem solved is a bit like installing a smoke alarm and assuming the house is fireproof.

The Window Is Still Open, But Not Wide

There are no confirmed, documented cases of a deepfake or AI-generated video successfully targeting a specific farm, named food processor, or agricultural operation as of early 2026. That’s worth saying clearly, because honesty is more useful than alarm. The threat in this sector is still precautionary. The playbooks being drafted are for something that hasn’t happened yet in this specific form, and I’d rather you prepare with accurate information than panic based on an inflated threat. That said, there’s still time to prepare, which is a luxury ransomware rarely extends.

The first wave of documented AI video manipulation around food hit food policy rather than food operations because of friction. Generating a fake video of someone ranting about food benefits requires zero knowledge of how any specific farm operates. No barn layouts. No equipment specifications. No breed knowledge. As tools mature and users get more fluent with prompt engineering, that friction drops. OpenAI’s Sora 2 app, launched in September 2025, generates short, highly realistic HD‑quality video clips (on the order of a few to tens of seconds) with synchronized audio and noticeably improved physical realism compared to the original Sora model. It hit a million downloads in under five days. In tests by NewsGuard, Sora 2 produced realistic videos advancing provably false claims 80 percent of the time (16 of 20 prompts) when asked to do so. Additionally, the visible watermarks can be removed with third-party tools in about 4 minutes, which is less time than it takes to microwave a frozen dinner.

The sector set the template for what happens when convincing visual evidence of abuse goes viral. The deepfake evolution of that template is not a distant scenario.

The Legal Situation (Spoiler: Agriculture Isn’t in It)

The legal landscape around deepfakes is moving. Food and agriculture are not in the conversation. (shocker…cough)The EU AI Act, which entered into force in August 2024, requires AI-generated content to be labeled in machine-readable format and deployers to disclose when content has been AI-modified. Transparency provisions take effect in August 2026, and fines reach up to 15 million euros or 3 percent of global annual turnover for violations. It applies to all sectors equally, which means fabricated farm footage is covered in the same breath as a deepfake politician, but there’s nothing tailored to the specific vulnerabilities of food system actors.

China moved first, with intense synthetic content regulations effective January 2023 requiring real-name registration for AI platform users, visible and metadata-level labeling of synthetic content, and 6-month log retention. It’s the most detailed framework of its kind anywhere, and, like the EU approach, it applies uniformly across sectors without acknowledging that a fake contamination video and a fake celebrity endorsement are not really the same kind of problem.

The U.S. TAKE IT DOWN Act, signed into law in May 2025, criminalized non-consensual intimate imagery, including AI-generated versions, and required platforms to remove flagged content within 48 hours. Important legislation for the people it protects. Completely irrelevant to a fabricated video of a chicken barn, which is not what it was designed to address.

Farms and food companies targeted by synthetic video can pursue defamation, product disparagement, and unfair competition claims under existing law, but those theories haven’t been tested in this specific context, and the creative lawyering required would take time that a viral crisis doesn’t offer. If your trade association has any appetite for legislative advocacy, getting deepfake-specific protection for food system threat actors onto an agenda now, before the first major incident forces a reactive scramble, is time well spent.

The Defense That Works Is Older Than AI

Here’s the part that I find interesting, and maybe a little satisfying: the most effective defense against an AI-generated attack on your operation is not an AI tool.

Deepfakes are good at generating plausible. They are considerably weaker at generating specific. AI video models are trained on broad visual patterns, and they flounder on hyper-local detail. Think of a very skilled set designer who has read about farms but never visited yours. They can build something that reads as a barn, but they’re going to get the gate latch wrong. The way your mud looks after three days of rain. The fact that your county doesn’t see snow before February. Your operation has a physical fingerprint that no model trained on generic agricultural footage can perfectly replicate. In a crisis, surfacing one or two of those mismatches is enough to create meaningful doubt about a fake, and meaningful doubt is all you need to buy time. The defense strategy is to document that fingerprint and hold it in reserve.

Before anything happens, build a baseline evidence kit. Time-stamped photos and short videos of key facilities, equipment, animals, and normal daily operations, taken regularly from multiple angles with location metadata intact. A shared folder accessible only to you and your crisis contacts. It doesn’t need to be sophisticated, but it just needs to exist. When something surfaces, you want to pull up dated documentation of “what your real operation looks like” and put it next to the fake within the hour, not the next morning.

Know your specific “tells” in advance. What would a fake video of your operation get wrong? Wrong breeds? Equipment you don’t own? A facility layout that doesn’t match? Colors of buildings or equipment? Seasonal details that don’t fit the timeline? Write them down. When you’re under pressure and your phone is ringing, you won’t want to be working through this from scratch.

Your verification network is as important as your documentation. Your vet, your field rep, your primary buyer, your commodity group contact, hell, even your neighbor, whoever knows your operation firsthand, these are the people who can say independently “that’s not their facility” when it matters. A third party saying that carries more weight than you saying it, and those relationships need to exist before the crisis, not be assembled during one. Calling someone you haven’t spoken to in two years and asking them to vouch for you publicly is an awkward conversation with poor odds.

Have a short holding statement drafted before you need it. A calm, factual acknowledgment that a video is circulating, that you are verifying its authenticity, and that your standards are what they’ve always been. One paragraph. Something that can go out within the hour to the press and your socials. The first window of viral spread is when containment is still possible, and scrambling to write under pressure while your inbox fills up is not the time to discover you have no idea what you want to say.

Tell your team not to argue in comments! Engaging with fake content in comment threads boosts its algorithmic reach and extends its life. Arguing with strangers on social media when you’re the subject of the attack is the digital equivalent of pouring accelerant on a fire because you’re hoping the fumes will smother it. Route everything to one person. Let that person respond once, clearly, with facts.

The guidance from agricultural communications professionals who specialize in deepfakes in animal agriculture is consistent: fold deepfake scenarios into your existing crisis communication plan. You already have crisis infrastructure. This fits inside it. Add the scenario. Run a drill. Update your contact trees.

The Trust Problem Is Bigger Than Any One Farm

I want to close with something that sits above any individual operation’s risk posture.

Each of the incidents I’ve described, the fake SNAP videos, the Indian street food deepfakes, the synthetic spokesperson risks, carries a damage mechanism that is harder to see than a lost retail contract. They degrade the shared evidentiary floor on which food system conversations depend. When enough synthetic content circulates convincingly, doubt becomes ambient. Consumers stop being able to reliably assess whether what they’re watching reflects reality. And once that skepticism spreads, it doesn’t just harm the targeted operations. It erodes the credibility of legitimate accountability footage alongside the fabricated stuff. It makes it easier to dismiss real evidence as fake and fake evidence as real, which is a terrible situation for anyone who cares about how food is produced.

Food systems run on trust, and it’s easy to take that for granted until it starts to crack. Trust is what makes a consumer reach for your product instead of the one next to it, what makes a retail buyer hold a contract when something uncomfortable surfaces, and what makes a regulator give you the benefit of the doubt during an inquiry. It isn’t soft or intangible. Its functional infrastructure and synthetic media are specifically designed to introduce fractures.

The food and agriculture sector has no deepfake-specific guidance from governments, no industry-wide activity addressing synthetic video, and no established industry-level response playbooks. That sits alongside concerning numbers: deepfake fraud drained $1.1 billion from U.S. corporate accounts in 2025, triple what it was in 2024, and the global volume of deepfakes grew from 500,000 in 2023 to 8 million in 2025. The sector’s combination of emotional charge, visual vulnerability, and current regulatory absence makes it a target worth taking seriously now, before the incident that changes the conversation.

The good news, and I mean this without irony, is that your physical operation is your advantage. The hyper-specific knowledge of what your barn looks like on a cold February morning, what your herd’s ear-tag sequence is, what your processing line sounds like at full speed, that is something no AI model can replicate without access to it, which it doesn’t have. You own that ground truth. Document it. Protect it. Know exactly how to use it when the moment comes.

Stay Safe, Stay Curious,

Kristin

📖 Pre-order my book Securing What Feeds Us: Cybersecurity in Food and Agriculture on Amazon. (If you want a different bookstore or an international link, please reach out.)

📬 Sign up for my newsletter, Food Systems IRL, and get updates on the book:

🎙️ Listen to the Bites and Bytes Podcast

🔗 Connect with me on LinkedIn