Anthropic Watermark: Claude Users Protest New AI System

Anthropic has elaborated on its watermarking system, which aims to comply with the European Union's Artificial Intelligence Act. This system makes it harder to claim AI-generated text as your own. What's interesting is that some Claude users are already canceling subscriptions in protest. An influencer named John Ennis posted a cancellation screenshot, citing the "ridiculous watermark idea." Many other users on X are also claiming to have canceled. Anthropic explains that the watermarking is not visible to readers. It works by subtly influencing token choices in the AI's output, especially in situations where multiple words could fit. For example, if the AI could choose "grey" or "overcast," the system might favor one based on a secret key. This key steers these low-stakes choices towards statistical preferences that are undetectable to humans. The technology is similar to Google's SynthID system. The bottom line is that while Anthropic aims for compliance, this new system is causing a strong reaction among some of its users.