AI Text Watermarking Is Free And Good

Don't Worry About the Vase
Read the full article
Scott Aaronson and Hendrik Kirchner developed a near-costless AI text watermarking method that subtly encodes outputs using a secret key-based pseudo-randomness source. Google has implemented this since 2024, and Anthropic recently adopted it to comply with the EU Code of Practice, enabling public detection without affecting output quality.

More on this story


More in AI Research

AI Research13 min read

What Is Data Poisoning? How Attackers Corrupt AI Models and Datasets

Unite.AI
AI Research1 min read

Shadow Computation Threatens CoT Monitorability

LessWrong
AI Research3 min read

Anthropic’s AI Hid a Secret Message in the Stars. OpenAI’s Bot Found It.

Gizmodo