// AI news · 2026-11-01 · The Decoder
Anthropic expands watermark to four Claude models with loophole
Anthropic extends watermark to four Claude models, including Opus 4.6 and Sonnet 4.6.
Published 2026-11-01 03:20 (Beijing time)Source The DecoderAuto-summarised, fact-checked against source
01Summary
The watermark is imperceptible, model-embedded, and platform-agnostic. On Opus 4.6 and Sonnet 4.6, it only activates at default sampling settings; minor parameter changes allow evasion. The method uses SynthID, a variant of Google DeepMind's technique, and applies across all hosting platforms.
Summarised from the original article by an open-source model, then checked automatically: every number and English name must appear in the source. The original article is authoritative.
02Key points
- Watermark extended to Claude Mythos 5, Opus 4.7, Opus 4.6, and Sonnet 4.6.
- On Opus 4.6 and Sonnet 4.6, watermark activates only at default sampling settings.
- Adjusting temperature, top_p, or top_k parameters allows evasion of the watermark.
- Watermark uses SynthID method and is embedded at model level across all platforms.
03Source
Anthropic's watermark rollout leaves a loophole on two older Claude models
The Decoder · published 2026-11-01 03:20 Beijing time
Also reported by 1 other outlet
05Related news
- Same source · OpenAI Updates AI Behavior Guidelines Emphasizing Correctability2026-10-31 · The Decoder
- Same topic · New AI-Powered Malware Volta-Autodoxer Discovered2026-10-31 · WIRED
- Same topic · AI Agent Performs Brain Injection in Live Mouse2026-10-31 · WIRED
- Same day · 12 Acceptance Tests for Swiss SME AI Projects2026-11-01 · Dev.to AI
From AIMap AI news: collected from 1,100+ official and media sources. Summaries only restate the original article; follow the link for the full story. All stories of 2026-11-01 · Archive by date · All AI news · Today