Anthropic Will Embed Machine-Readable Marks in Claude Outputs Worldwide

Overview

• Anthropic announced that texts from Claude will carry an invisible, machine-readable watermark and generated images will include signed provenance metadata using the open C2PA standard to record origin and tamper history.
• The watermark is designed to be imperceptible to readers, to travel with copied text and to persist after light edits, while the signed image metadata will record creation details such as format and date.
• The company cautions the mark only signals that content was processed by Claude and is not conclusive proof of sole authorship because the platform can mark material it only reviewed, translated, summarized, or converted.
• Models released after August 2, 2026, will include the marking and Anthropic plans global rollout where Claude is offered, while older model versions will need updates and a transition period with no public timetable.
• Anthropic says it will provide detection tools and guidance for users and developers but has not released technical details or dates; the company is also reported to be in early talks to buy Israeli AI startup Decart as it expands infrastructure.