Tech Giants Sign EU Code, Claude Pioneers Invisible AI Text Watermarks
机器之心 · wechat · 2026-08-13
Major AI companies including OpenAI, Anthropic, and Google have signed the EU Code of Practice on General-Purpose AI, committing to advancing the labeling and detection of AI-generated content.
Anthropic's Implementation for Claude:
- Text Watermarks: Embedded directly into generated text. Invisible to humans but machine-detectable, surviving copy-pasting and light editing.
- Image Metadata: Appends digitally signed C2PA metadata to image files to prove origin and detect tampering.
- Scope: Applied at the model layer for all users globally (not just the EU). New models already have it; older models must comply by December 2.
Technical Background & Controversy:
Google DeepMind previously deployed SynthID-Text, which subtly adjusts token probabilities to create statistical patterns. While tests show no degradation in text quality, the public remains skeptical about the watermark's robustness against heavy rewriting and the potential shifting of fact-checking responsibility onto readers.
More from Models
- First Model to Pass the Video Turing Test: Half of Users Thought It Was Human — Puzzleheaded-King584 · 2026-10-03
- Using System One models in Swift: fast, deterministic decisions via Apple Foundation Models — rxwei · 2026-10-03
- Linux Kernel CVEs Surge From ~500 to 1500+ Per Release, LLMs Blamed for Bulk of the Rise — burny_tech · 2026-10-03
- Developer Complains OpenAI's Coding Model Endlessly Scopes Creeps Instead of Finishing Tasks — DavidWells · 2026-10-03
- Sonnet 5 Spotted in Google Antigravity Backend, Which Still Runs Sonnet 4.6 — brandon_galang · 2026-10-03
- Post-training Yandex AliceAI-80B-A3B from scratch: a NaN bug in custom V100 kernels killed one run — jjusko20 · 2026-10-03