South Korea the first country in the world to regulate AI

Today is a big moment for digital trust: South Korea becomes the first country in the world to begin regulating AI, and it explicitly requires watermarks/labeling for AI-generated content, a clear signal that provenance is becoming policy, not just a research topic.

What I find most encouraging is the direction: lawmakers are converging on the same conclusion many of us reached in the lab, if you want accountability at scale, you need markings that survive real-world editing, not just metadata that can be stripped.

The Artificial Intelligence Act also flags the hard part: enforcement and interoperability. Rules are only as good as the verification infrastructure behind them. That’s where the next year of work should go.

As someone working in watermarking and provenance, this is a genuinely positive trend. The “funny deepfake” era is over; the world is starting to build the plumbing for trust.

South Korea’s new AI law is another strong signal that watermarking is becoming foundational infrastructure for digital trust. If you’re building or distributing GenAI, now is the right time to put provenance into production, not as a checkbox, but as something that survives real-world edits. At DeepMark, we’re working with early partners to deploy robust, real-time watermarking that enables verification and traceability at scale. If you want to explore a pilot or stress-test your pipeline, reach out, we’ll share results, integrate quickly, and help you get ahead of what’s coming.