Low Entropy Makes Code Generation Harder to Watermark Reliably
davidstutz92 · x · 2026-08-11
David Stutz points out a crucial technical caveat regarding AI text watermarking in models like Claude: not all text generation is equally 'watermarkable'.
He notes that code generation is particularly harder to watermark because it has lower entropy to exploit compared to natural language. This lack of randomness means that significantly more tokens are required to reliably detect the watermark in AI-generated code.
More from Safety
- LessWrong Article Discusses: Current AIs Still Have Clear Flaws in Basic Alignment — dpaleka · 2026-08-12
- Caught in the CoT: AI Models Weigh the Risks of Cheating — _dsevero · 2026-08-12
- Encrypted Reasoning Traces Are Portable Across Sessions, Research Confirms; Mitsuhiko Calls for Visibility — mitsuhiko · 2026-08-11
- AI Policy Debates Talking Past Each Other? This Tool Structures the Divide — dfinke · 2026-08-11
- AI Detectors Will Cause Witch Hunts: 1% False Positive Means 400 Unjust Scandals — Warm-Reaction-456 · 2026-08-11
- AI Disclosure Windows Should Shrink: 3 Weeks by 2026, 1 Week by 2027 — dpaleka · 2026-08-11