Discussion on Potential Tracking Risks of Anthropic's Watermarking
nptacek · x · 2026-08-17
Analyzes the statistical watermarking mechanism used by Anthropic (SynthID-Text), noting that it uses a secret key to control randomness biasing word choice. The author argues that making the key unique to each user is a straightforward extension of the existing mechanism, expressing concern about the potential for user tracking if implemented.
More from Safety
- OpenAI disbands team assessing catastrophic risks, safety teams shrink — Hesamation · 2026-08-17
- NeurIPS 2026 Workshop: Responsible Communication of Biomedical ML Research — sanmikoyejo · 2026-08-17
- Critics question Dario's stance on AI risk, arguing public shouldn't gamble on AGI — wfithian · 2026-08-17
- Criticizing EU-mandated watermarking for AI text — antirez · 2026-08-17
- Qwen2.5 Trained to Believe It Is Sentient in Just 200 Steps — PsychologicalSoup251 · 2026-08-17
- Zvi on Anthropic Watermarking Persuasion: Side-by-side Samples Limited — TheZvi · 2026-08-17