Protecting Content From AI

AI didn’t break content. It exposed something that’s been broken for a long time.

For years, digital content has moved through the internet stripped of authorship, context, and intent. We normalized it. Files were copied, transcoded, reposted, scraped—metadata lost at every step. Then AI showed up and started training on the wreckage. The result wasn’t surprising. We just finally noticed the consequences.

What’s missing isn’t regulation, or ethics statements, or stronger lawsuits. What’s missing is structure.

That’s what provenance actually is. Not enforcement. Not DRM. Provenance is the ability to answer, with certainty, who created a piece of content, what’s happened to it since, and what obligations travel with it. Without those answers, everything downstream—platforms, partners, and now AI models—operates blind.

This is where C2PA matters. It turns metadata into something verifiable. Authorship, edits, tools used, and intent aren’t just claims anymore; they’re cryptographically signed and tamper-evident. That changes the conversation. Once provenance exists, ambiguity disappears. And ambiguity is what AI systems have been living off of.

Take “Do Not Train.” Most people talk about it like it’s a technical block. It’s not. It’s a notice mechanism. Its power isn’t in stopping ingestion; it’s in removing plausible deniability. When a model trains on content that carries a signed “Do Not Train” assertion, that’s no longer an accident. It’s a decision. C2PA makes that intent durable, portable, and provable.

Copyright has the same problem. The law assumes provenance. The internet removed it. AI just scaled the gap. Without a verifiable creator, timestamp, and chain of custody, copyright becomes theoretical. With provenance, it becomes evidence. Courts don’t rule on vibes. They rule on proof.

The uncomfortable truth is that metadata is now the control plane for content. If your metadata can’t survive distribution, you don’t actually control your assets. If it isn’t verifiable, no downstream system—human or machine—has any obligation to respect it.

That’s why we built Limbo.

Limbo is provenance infrastructure for the real world. It helps enterprises validate and generate C2PA metadata, preserve intent through transformations, and clearly signal things like authorship, copyright, and “Do Not Train” without rewriting their entire stack. It’s not about stopping AI. It’s about making trust, responsibility, and liability explicit again.

AI will keep accelerating. That part is inevitable.
What’s optional is whether your content continues to exist without structure.

If you care about protecting what you create, provenance isn’t a nice-to-have anymore. It’s the foundation.

You can see how it works at trylimbo.com.

‍