You scroll through your feed. A video of a politician saying something outrageous goes viral. An article quotes a CEO making absurd promises. Is it real? Is it AI? You have no idea. And the people building this technology claim they have a solution: watermarks.
Anthropic just pledged to embed watermarks into its AI-generated content to appease EU regulators. It sounds like a win for accountability. It feels like someone is finally taking the spread of synthetic media seriously. But it’s a lie.
AI watermarks aren’t a shield against disinformation; they’re a liability transfer.
Think about it for more than two seconds. Removing a digital watermark is a trivial task. Any bad actor with basic technical skills can strip it out before republishing a deepfake or a flood of AI-generated propaganda. The regulators know this. Anthropic knows this. Yet they proceed as if plastering an invisible “Made by AI” tag on a file will save democracy.
Why? Because the alternative—actually enforcing detection at scale—is politically unpalatable and technologically brutal. It’s much easier to engage in regulatory theater.
When the solution is technically useless, you can bet it’s actually a legal weapon.
This isn’t about protecting you from AI slop. This is about creating a paper trail for blame. By embedding a watermark, Anthropic gets to claim compliance. The EU gets to claim they took action against Big Tech. And when an AI-generated video ruins someone’s life or disrupts an election, who gets blamed? Not the multi-billion dollar model maker. Not the government.
The blame falls squarely on the downstream user who stripped the watermark. The legal framework shifts the target from the creators of the AI to the consumers and sharers of the content. You become the scapegoat.
We are being sold a narrative of control. We are being told that a nagging, easily bypassed tag will restore trust to the internet. But this strategy doesn’t fix the problem of synthetic media; it just weaponizes the legal system against the people consuming it.
They aren’t protecting you from the machine; they’re protecting themselves from your blame.
The next time you see a regulator boasting about AI watermarking, don’t sigh in relief. Get angry. Because they aren’t solving the crisis of trust. They are just making sure that when the internet finally drowns in AI slop, you’re the one holding the bag.
FAQ
Q: Aren't watermarks better than nothing at all?
A: No. A false sense of security is worse than no security. It creates the illusion of compliance while actively shifting the legal risk onto you.
Q: What does this mean for how I consume content?
A: It means you still cannot trust anything you see online, even when regulators claim AI is 'tagged.' You remain the one bearing the risk.
Q: Why would Anthropic agree to a technically useless measure?
A: Because performative compliance is cheap. It appeases politicians, protects their valuations, and leaves the perfect scapegoat for future lawsuits: the end user.