Back to all articles
•Technology

Anthropic Introduces Claude's AI Text Watermark to Comply with EU Regulations

View original source

Anthropic has announced plans to implement text watermarking for its chatbot Claude in compliance with the EU AI Act’s Transparency Code. The watermarking process, aimed at distinguishing AI-generated content, has sparked controversy among Claude users. Some view it as a conspiracy against users, while others see it positively as a step towards increased transparency.

  • The watermark uses a technique called SynthID-Text, developed by Google DeepMind.
  • The watermark is said to be undetectable in everyday reading but identifiable with a specific key by Anthropic or other authorized parties.
  • Some users express concerns about privacy and potential misuse, as referenced by some canceling subscriptions.
  • Rewriting or heavily modifying text can potentially remove or alter the watermark.
  • The impact on the code generated by Claude is expected to be minimal, except in variable or comment choices within the code.
  • Other major AI companies are expected to follow similar practices due to shared compliance commitments.