Why OpenAI Won’t Watermark

OpenAI has had a working text watermark for ChatGPT for over a year. The Wall Street Journal reports it works exceptionally well 99.9% effective, resistant to light editing, and without any noticeable drop in output quality.

Yet they haven’t released it.

The official explanation is equity. OpenAI points out that watermarks might unfairly harm non-native English speakers who rely on AI to edit their work. There is some truth to this, but it misses the main point.

The real reason is simpler: nearly 30% of surveyed ChatGPT users said they would use the tool less if watermarks were added.

When a company builds something that works, but refuses to launch it because a third of its users would leave, that isn’t a technical problem. It’s a fundamental conflict between what a product is useful for and what its creators want to be seen doing.

Right now, OpenAI is caught in that split. Educators want a tool to stop cheating; users want a tool that can’t be traced; OpenAI wants both to like them. So instead of deploying the system they built, they are exploring unproven alternatives like cryptographic metadata hoping to find a clever technical trick that lets them avoid making a hard choice.

In tech, when you delay a working solution to wait for a perfect one that pisses no one off, you rarely end up with better technology. You just end up waiting.