Introduction:

In June 2026, Google made a legal argument that should stop every creator in their tracks. Google was facing a copyright lawsuit over its Lyria 3 AI music model. Google didn't argue fair use. It argued something simpler, and more unsettling. Independent musicians who uploaded songs to YouTube had already agreed to AI training. Years ago. In the terms of service they clicked through just to post a video.
The clause Google pointed to is from 2019. It never mentions AI. It never mentions training. It was written to cover ordinary hosting and streaming. Google's position is that none of that matters. The license was broad. Broad is broad, according to Google, no matter what it was written for.
That case is still unresolved. But it's the clearest real-world proof of Retroactive Consent this series has found. Old terms, stretched to cover a use nobody named at the time. It's the legal strategy of one of the largest platforms on Earth, made in a real courtroom, this year.
Here's what YouTube, TikTok, and Meta's terms actually allow. Broken into what stays inside the company, and what can leave it.

YouTube and Google

Inside the company: YouTube's terms give Google a broad license to use what you upload. That includes making new versions of it. The license applies to Google's business generally, not just YouTube. YouTube's own CEO has said some YouTube videos get used to train other Google AI models too, including Gemini.
Leaving the company: The same license can be passed to other companies. It's not limited to Google using your content in-house. The clause is old and general. It was written before generative AI existed. It never names AI or training as a use. That's exactly the part Google now says doesn't matter.
What you can't undo: There's no real opt-out in the standard terms. If the court agrees with Google's reading, anything you already uploaded is fair game. Not just content posted after AI became a stated use.

TikTok

Inside the company: TikTok's terms are the most direct of the three. They plainly state that content can be used to build and improve TikTok's own AI models. The words “training” and “testing” are right there in the terms, not just implied.
Leaving the company: The license reaches beyond TikTok itself. It extends to its parent company, its service providers, and its business partners. Content can also go to verified marketing partners for promotion.
What you can't undo: TikTok is clear that there's no opt-out for public content. Switching a video to private afterward doesn't pull it back out of a model that already learned from it. Deleting your account doesn't erase the training use either.

Meta: Facebook, Instagram, and Threads

Inside the company: Meta uses public posts, photos, and captions to train its AI systems by default. It relies on something called “legitimate interest” as its legal basis. That means it doesn't ask for your opt-in agreement first.
Leaving the company: Meta's license can extend to advertisers and partners, not just Meta. Meta's AI tools also share some user inputs with select outside companies to help generate responses.
What you can undo, partially: Meta is the only one of the three with a real opt-out. EU and UK users can file a formal objection under GDPR, and Meta must honor it. US users can file the same objection, but Meta isn't required to act on it. There's also a content-reuse toggle in Instagram settings. It stops future reuse, but does nothing about content already used.

What These Rights Actually Enable

A license clause is easy to underestimate when you only read it in the abstract. Here's what it enables in practice.
Voice cloning, from any video where you're speaking on camera. Face and likeness generation, the same way Meta's Muse Image tool works. Deepfake video and synthetic endorsements, using your face and voice together. Third-party resale, meaning the platform doesn't have to use your content itself. It can hand usable rights to someone else. Training data for products that have nothing to do with the platform at all.

The Pattern

The same four things repeat across all three platforms. The license is broad by design, written in language flexible enough to cover uses that didn't exist yet. Consent is the default, not a real choice. Just continuing to use the platform is treated as agreement. The license usually isn't limited to internal use, thanks to words like “sublicensable” and “transferable.” And deletion doesn't reach backward. Removing a post usually stops future use, but doesn't undo training that already happened.

What To Actually Do

Check your account settings on each platform for a content-reuse or AI-training toggle. Instagram has one, though it isn't retroactive.
File the objection form if you're in the EU or UK, since it carries legal weight there. File it in the US anyway. It still creates a documented record.
Assume anything you've posted publicly on any of these platforms may already be part of a training set. The decision that still matters is what you post going forward.
Watch for the words “sublicensable” and “transferable” in any platform's terms. That's the difference between a company using your content, and a company being able to hand rights to your content to someone else entirely.

Where This Fits

This is one piece in a larger series about Retroactive Consent, the pattern of old agreements being stretched to cover new AI uses. The Lyria 3 lawsuit is still being watched closely. If Google's argument wins, it sets a precedent well beyond music, and well beyond YouTube.

Where This Goes Next

Retroactive Consent isn't limited to platform terms of service. The same structure — old terms, stretched to cover a use never named — shows up anywhere a platform, a vendor, or a client-recording tool sits between a person and their own voice, face, or written work. This series will keep applying that same pattern, case by case, to:
● Voice actors, audiobook narrators, and broadcasters — the same BIPA suits referenced in this series, examined through each profession's specific contracts and exposure.
● Face-forward creators — fitness, cooking, and lifestyle creators whose likeness is the product itself.
● Coaches, consultants, and service providers — where the exposure isn't personal biometrics but client data and vendor liability.
● Musicians, authors, and visual artists — where the same argument Google is making about a 2019 YouTube clause is already being tested against record labels, publishers, and stock-photo archives.
Each of those segment breakdowns will follow the same discipline this one does: state plainly what's confirmed, what's attributed to a source rather than independently verified, and what remains an open question — and end with what actually helps, not just what to be afraid of, so you're clear on what action you can take if it doesn't sit well with you.

What's Your Next Move

Want to continue conversations like this with other entrepreneurs, founders, and leaders?

Join the Majority Media communities → https://community.majoritymedia.us/

Ready to work with me? → https://seanmatkinson.com/start-here/
Subscribe to The Strategic Thinkers Podcast to catch the rest of the series.