Particle.news

Anthropic Embeds Machine‑Readable Marks in Claude Outputs

It aims to help platforms and regulators spot AI‑generated content despite detection methods not yet published.

Overview

  • Anthropic began marking Claude outputs to meet Article 50(2) requirements, implementing the change on August 2, 2026.
  • Text from supported Claude models will carry an imperceptible, machine‑readable watermark that Anthropic says does not change meaning, quality, or readability.
  • For supported files such as .svg, .png, and .jpg Claude will attach digitally signed provenance metadata that follows the C2PA open standard.
  • The marking is applied at the model level across Claude products worldwide and Anthropic has committed to transitional marking support for models launched before the enforcement date.
  • Anthropic has not published full technical specifications or detector tools yet, leaving questions about watermark robustness, persistence through edits, and practical detection workflows for third parties and regulators.