Smile For The Algorithm

YouTube built something fun. The lawyers already knew what it was.

Part of the ongoing Big Tech's War on Users series.

YouTube announced something fun yesterday.

No really. Genuinely fun on the surface. You can now create an AI avatar of yourself for use in Shorts. Record a quick selfie, read a few prompts so it can capture your voice, and congratulations — you have a photorealistic digital double that can star in your videos without you having to actually be there. Convenient. Creative. Very now.

The feature is rolling out globally starting yesterday. Outside Europe, but we'll get to that.

Google says it's safe. Your avatar can only be used by you. Nobody else can take your digital face for a spin. It'll be watermarked. Labeled. Handled responsibly. They even delete it after three years of inactivity, which is the kind of reassurance that sounds thorough right up until you think about it for thirty seconds. Maybe twenty.

So. Fun feature. Great. Cool. Fine.

Smile for the camera.

But.

It's not actually a YouTube feature.

Look at the setup instructions. To create your avatar you tap the Gemini spark. That's not a YouTube button wearing a Gemini costume. That's a Gemini feature that lives inside YouTube. Which means you're not handing your face and voice to a video platform. You're handing it to Google's entire AI ecosystem, which Gemini is increasingly the connective tissue of.

That changes the sentence considerably.

And Google's AI ecosystem has a privacy policy that states — not in the fine print, just in the policy, sitting right there — that public content, including biometric information, can be used to train Google's AI models. YouTube says they would never do that with your avatar data. They're reviewing the language to make it clearer.

They are not changing the policy.

So we have a biometric data collection feature built on infrastructure governed by a policy that explicitly permits using that data for AI training, and a pinky promise that they won't. Experts who spoke to CNBC were not reassured. Two of them said they wouldn't recommend their clients sign up.

But sure. Fun feature.

Now here's where it gets interesting.

To use this feature you have to be logged in. Which is obvious. Of course you do. How else would YouTube know it's your avatar.

But let's think about what being logged into YouTube actually means in 2026.

Your Google account isn't a YouTube account. It's everything. It's your search history. Your Gmail. Your Google Maps location history — where you've been, when, how often, whether you drive past the same medical building every third Tuesday. Your Chrome browsing data if you use it. Your Google Photos, which has been quietly doing facial recognition on your uploads for years. Your Android device data if you're on that ecosystem. Your Play Store purchases. Your Google Pay history if you use that.

And now your face. Your voice. Photorealistic. User-verified. Clean source material that you quality-checked yourself by doing the setup process.

This isn't a YouTube profile anymore. This is the most complete consumer identity dossier ever assembled, and you just stapled your biometrics to it. Voluntarily. Because the feature looked fun.

And before anyone reaches for the "I turned off my activity tracking" comfort blanket — that helps less than Google's settings menu implies. The data Google holds on you isn't purely a function of what you've consciously left on. It's behavioral. Inferred. Built from signals you didn't know you were sending. The way you search. The way you scroll. The content you linger on half a second longer than the content you don't. The device fingerprint. The network. The pattern of when you show up and when you don't.

Turning off the memory doesn't erase the model they've already built. It just stops updating it in the ways they tell you about.

You were already a profile before you touched the avatar feature. You just handed that profile a face and a voice to go with it.

The ad targeting implications alone should give you pause. Google isn't just the world's largest search engine. It's the world's largest advertising machine. That's the actual business. Everything else is infrastructure for that. And they just added something to your profile that ad targeting has never had before — not a probabilistic match, not a third party cookie, not an educated guess based on behavior.

You. Confirmed. By your own face and voice.

The advertiser doesn't need to wonder if you're the same person who searched for that thing last Tuesday. They know. You told them. You smiled and read the prompts and handed it over because you wanted to make a fun video.

Which is when the other shoe drops.

This also solves something platforms have been fighting regulators about for years. Age verification. The debate about whether platforms should be required to verify user ages is a massive ongoing legal and political mess. Governments legislating it. Platforms resisting it. Privacy advocates arguing about implementation. Nobody agreeing on anything.

Google just walked around the entire debate.

You didn't get asked to verify your age. Nobody demanded your ID. No controversial mandatory verification. Instead you opted into a fun creative feature that happens to require the same information age verification would have required anyway. Account in good standing. Biometric face and voice data. Identity confirmed.

They didn't need to win the regulatory argument. They just needed a cool enough toy that you'd run toward the finish line yourself.

Accidental? Maybe.

But then again. They didn't launch this in Europe.

This is the same platform that has spent 2026 pulling channels for producing too much low-effort AI content. The policy says less AI slop. The product says here's a tool to generate photorealistic AI versions of yourself at scale. Both things are true simultaneously.

There's another layer that the launch coverage almost entirely missed.

The safety framing — your avatar can only be used by you, watermarks, SynthID, C2PA labels, very responsible, very thorough — assumes the content stays on YouTube's platform. Under YouTube's rules. Subject to YouTube's controls.

It won't.

People have been downloading YouTube content since approximately fifteen minutes after YouTube launched. YouTube-dl. yt-dlp. Browser extensions. Screen recorders. This is decades-old behavior so deeply embedded in how people use the internet that it's essentially a parallel infrastructure at this point. YouTube has never successfully stopped it. They've tried. It keeps working anyway.

So the moment your avatar appears in a Short, it's downloadable. And what gets downloaded isn't a fun video. It's clean, high-quality, photorealistic footage of your face and your voice — source material that you personally produced, quality-checked, and published. Better material than most bad actors could scrape together on their own.

Feed that into any of the less scrupulous open source or offshore AI tools — and there are plenty, some pulled from app stores, some operating in jurisdictions that don't particularly care about your content policies — and YouTube's guardrails become completely irrelevant. They have no jurisdiction over what happens outside their platform.

This isn't a hypothetical. We've watched this exact cycle play out repeatedly. There's a graveyard of tools that tried something similar and got shut down or kneecapped precisely because of the abuse potential. Reface had to add restrictions after the obvious misuse cases showed up immediately. Avatarify got neutered after it blew up. FaceSwap apps disappeared from app stores entirely. Lensa AI's Magic Avatars exploded in controversy over what it did with uploaded photos. MyHeritage's Deep Nostalgia got heavily restricted. And Sora — which actually had an almost identical avatar feature called cameos that let users scan their faces to create digital doubles — lasted six months before getting shut down, with its watermarks cracked within weeks of launch.

The pattern is always identical. Tool launches. Safety team says guardrails are in place. Community spends roughly 72 hours finding ways around them. Company either restricts the feature or quietly disappears it.

The difference here is scale and infrastructure. Those tools were trying to build a user base from scratch. YouTube already has two billion of them. And unlike those tools, this one isn't a standalone product that can be pulled from an app store.

Remember — Google owns one of the two major app stores. And Apple, the one platform that could theoretically push back, just signed a deal to bring Google's AI into Siri, on top of a search revenue relationship estimated in the tens of billions annually. Nobody is pulling this app. Nobody is coming to remove the feature. The usual mechanisms that eventually reined in the smaller players don't apply here.

Which means the source material — your photorealistic face, your cloned voice, purpose-built and user-verified — is going to be out there. Downloadable. Usable. Permanently.

You can delete your avatar. YouTube will even do it automatically after three years of inactivity.

What you cannot delete is what was already learned from it. What you cannot recall is the footage already downloaded. What you cannot un-ring is the bell of having handed a bad actor better deepfake source material than they could have assembled themselves, wrapped in a consent form you tapped through to make a fun video.

Google's safety framing didn't prevent the risk.

It just made the source material better.

Now. The EU.

YouTube is rolling this out globally. Starting today. Outside Europe.

That's a business decision that requires deliberate technical implementation. You don't accidentally exclude 450 million people. Geofencing a feature at launch means someone made a list of where it would and wouldn't be available, and Europe was a conscious entry on the wrong side of that list.

The reason sitting underneath that decision has a name. GDPR. Specifically Article 9, which treats biometric data used to uniquely identify a person as a special category requiring explicit legal justification beyond standard consent. The bar is considerably higher than "user tapped agree on a fun avatar screen."

Which means somewhere inside Google there are documents. Legal memos. Data Protection Impact Assessments. Risk evaluations. Privacy counsel sign-off. That explicitly lay out why this feature cannot launch in the EU in its current form.

Those documents describe the implications.

The biometric profiling. The account identity locking. The AI training policy gap. The age verification angle. The Gemini attack surface. Possibly all of it. Possibly more than we've covered here.

They know. The lawyers know. The product team knows. The executives who approved the rollout geography know.

So the "we didn't think about it" defense — which to be clear nobody has made yet because nobody's asking the right questions yet — dies right here. The EU exclusion is the timestamp on the document that proves otherwise. You cannot simultaneously know enough to protect yourself legally in one jurisdiction and claim ignorance about the implications everywhere else.

What the EU exclusion actually tells you is that Google looked at exactly what this feature does, weighed it against a regulatory framework with actual teeth, and concluded the risk wasn't worth it there.

Then launched everywhere else anyway.

Worth noting: Google is currently under EU investigation for whether it imposes unfair terms on creators for AI. They launched a biometric avatar feature outside Europe the same week. Make of that what you will.

Your face. Your voice. Your profile. Your problem.

They'll get to Europe eventually. After the lawyers figure out the minimum concessions required to satisfy regulators while preserving as much of the architecture as possible. That's the other half of launching outside Europe first — it's not just avoidance. It's a testing period. Hundreds of millions of enrollments to establish the product, build the case for its benefits, generate the user satisfaction data, and present regulators with a fait accompli wrapped in a consent form.

By the time Europe gets the feature it'll be dressed differently. Same infrastructure underneath.

And while we're watching spaces, keep an eye on your iPhone.

Memoji exists. It's normalized. It's fun. It's very Apple. And it currently feels safe because it's local, it's Apple's ecosystem, and Apple has spent fifteen years and billions of marketing dollars on the privacy brand.

Then Apple signed the Siri deal with Google.

So here's the what if. What if a photorealistic Memoji upgrade shows up powered by the same avatar pipeline? What if the feature looks like Apple but the engine is Google's? The users least likely to see it coming are precisely the ones who made the most deliberate choices to stay out of Google's ecosystem. They picked the phone with the privacy reputation. They turned the tracking off. They thought they were out of the blast radius.

And Apple will tell you the infrastructure is separate. Custom implementation. Walled off. Today that may even be true. But infrastructure separation is an agreement, not a law. One renegotiation. One API update. One quietly amended term between two companies whose financial relationship gives Google considerably more leverage than your privacy preferences do.

The wall is maintained by paperwork and good intentions.

We've seen how durable those are.

That thread deserves its own post. And it'll get one. But consider the flag planted.

For now let's bring it back to today's announcement. Because the Google and Apple entanglement is a whole separate post — and frankly it deserves one.

Today is about YouTube. Today is about a fun feature. Today is about smiling for the algorithm.

Here's what you actually did if you signed up this morning:

You handed Google a photorealistic copy of your face and a clean verified sample of your voice. You tied it to an account that already holds your search history, your location data, your email, your browsing patterns, your purchase history, and years of behavioral inference that persists regardless of what your privacy settings say. You became the most precisely targeted advertising subject Google has ever had. You volunteered the exact information that age verification legislation has been trying to force platforms to collect for years. You created better deepfake source material than most bad actors could have assembled on their own. And you did all of it by tapping through a consent screen to make a fun video.

The lawyers already knew. The EU exclusion is the receipt.

Everyone else got the feature.

So the next time you're scrolling Shorts and the prompt appears — and it will, algorithmically timed to the moment you're most likely to tap it — ask yourself one question before you hit that button.

Not "is this safe." Google has people whose entire job is to make sure that question leads somewhere comfortable.

Not "will anyone misuse this." That ship has historically sailed approximately 72 hours after every similar feature ever launched.

Just this: who benefits?

The feature is free. The toy is fun. The avatar is yours.

And you are the product. Photorealistic now. Voice-verified. Tied to everything they already know about you. Delivered voluntarily with a smile.

They didn't even have to ask twice.

Smile for the algorithm.