Deep Dive: AI Text Watermarking Is Free and Good, So Why Is Everyone Mad?

Don't Worry About the Vase (Zvi) · rss · 2026-08-22

Zvi provides an in-depth analysis of AI text watermarking. The method, pioneered by Scott Aaronson, uses pseudo-randomness to embed a signal in outputs with near-zero marginal cost and no perceptible impact on humans. Google and Anthropic are implementing it to comply with EU regulations. The article argues that the public backlash stems largely from bias against Anthropic, misunderstandings of LLM stochasticity, and fear of being caught, rather than technical downsides.

Original post →

More from Safety

Safety channel →