Google's Veo 3 Is Astonishing Technology, and That's Exactly the Problem

Ever since OpenAI unveiled Sora back in 2024, the arrival of generative AI video has felt inevitable, dangling the promise that anyone could one day conjure realistic footage from a line of text. Google has now jumped into that race with a genuinely impressive Gemini tool, one that also happens to be capable of manufacturing disinformation on a terrifying scale.

I spent time putting Google's heavily promoted new video model, Veo 3, through its paces. It ships inside Gemini's steep $250-a-month AI Ultra tier, and it can render tiny, intricately detailed objects in motion, say, a pile of chopped onions, alongside convincing synthetic audio. It isn't flawless. But with patient prompt tuning and enough attempts, Veo 3 will hand you clips that, at a casual glance, are impossible to distinguish from real life.

This is undeniably slick, seriously impressive engineering. It's also a lot more than that. It may well be the final nail in the coffin for truth online. Veo 3 is already a serious threat as it stands, and a single modest update could turn it into a weapon for deepfakes, online harassment, and mass misinformation.

The Moment Veo 3 Gets Image Uploads, the Game Is Over

For all the ways Veo 3 leapfrogs its predecessor, Veo 2, it's currently missing one pivotal capability: generating video from a photo you supply.

With Veo 2, I can feed in a picture of myself and have it spin up a clip of me hunched over my computer. Given that both Veo 2 and Google's Whisk animation tool already support this, it feels like a foregone conclusion that Veo 3 will eventually get it too. (We've reached out to Google to ask whether that's the plan and will update this piece if they respond.) Once that lands, anyone will be able to fabricate lifelike video of people they know saying and doing things they never said or did, and never would.

The consequences write themselves in a moment when clips of questionable origin already tear through social feeds daily. Annoyed at your manager? Fire off a clip to HR of them behaving badly. Looking to seed fake news? Drop a counterfeit press conference onto Facebook. Resent an ex? Generate them doing something humiliating and ship it to their whole family. The only real ceilings here are your imagination and your conscience.

If producing a video, with audio, of a real person comes down to a handful of clicks and costs little or nothing, how many people will misuse it? Even if only a sliver of users do, that still adds up to enormous potential for havoc.

Google Isn't Taking Moderation Seriously

Unsurprisingly, Google places some guardrails on what Gemini will and won't do. But the company is nowhere near strict enough to head off the worst outcomes.

Out of every chatbot I've tested from the major tech players, Gemini has the flimsiest restrictions. It's not supposed to traffic in hate speech, yet it'll cough up examples if you ask. It's not supposed to make sexualized images, yet prompt it and it'll render someone in beachwear or lingerie. It's not supposed to facilitate illegal activity, yet ask and it'll rattle off the top torrent sites. Token barriers that stop Gemini from generating, say, a video of a well-known politician simply aren't enough when its policies are so trivial to sidestep.

So what happens when Google's already-loose rules collide with an online crowd determined to break them? Look at ChatGPTJailbreak, a subreddit ranked in the top 2% by size. The community is devoted to "unlocking an AI in conversation to get it to behave in ways it normally wouldn't due to its built-in guardrails." Picture what people of that bent will do once they get their hands on Veo 3.

I couldn't care less if someone wants to entertain themselves by coaxing a chatbot into adult content or using it to hunt down torrents. What worries me is what easily generated, photorealistic video, complete with sound, means for harassment, misinformation, and public discourse.

Living With Veo 3's New Normal

For every SynthID watermarking scheme Google rolls out, a crop of third-party removal sites and how-to guides springs up to defeat it. For every chatbot fenced in with safeguards, there's a FreedomGPT that ditches them. Even if Google clamps Gemini down so hard you can't squeeze out a harmless cat video, almost nothing stands in the way of jailbreakers and uncensored copycats once Veo 3-class generation goes mainstream.

For decades, crude Photoshopped images of real people doing things they never did have circulated online, just a fixture of digital life. By the same logic, you have to fact-check anything you encounter that seems too awful or too good to be true. That's the new reality with Veo 3-grade video: you can no longer assume any clip is genuine unless it comes from a credible news outlet or another source you have real reason to trust.

And Gemini's Veo 3 is merely the first stone skipping across the pond of widely available, truly convincing AI video. These models will only grow more realistic, accumulate more features, and spread further. The era when video footage of an event served as the smoking gun is ending. Truth isn't dead, exactly, but it's a different thing now, and it demands careful verification.