A Mark You Cannot See, but Can Measure

Anthropic announced on August 13, 2026 that it will start embedding an invisible watermark in text generated by Claude models that support the feature. According to the company's official announcement, confirmed by Forbes and Gizmodo, the mark works by subtly shifting the model's word choices during text generation, a pattern imperceptible to readers but statistically identifiable once enough text is available to analyze.

Based on Research From Rival Google DeepMind

The technique Anthropic uses is an adapted version of SynthID-Text, a watermarking approach Google DeepMind published in a Nature paper in 2024, according to BleepingComputer's reporting. The mark does not change the meaning, quality or readability of Claude's response, and it remains present even when the text is copied and pasted into another document or system.

Where the Mark Already Applies, and Where It's Headed

According to Anthropic, Claude models launched in the European Union on or after August 2, 2026 already ship with support for the marking system. The feature's rollout covers the Claude Platform, used via API, Claude.ai, Claude Code, Claude Cowork and Claude Tag, spanning nearly the company's entire product line in any country where Claude is available.

Images Also Get Signed Provenance Metadata

Beyond text, Anthropic is now attaching digitally signed provenance metadata to image files generated in formats such as .svg, .png and .jpg. The company thus joins OpenAI and Google, which had already publicly detailed how they intend to comply with the European Union's AI Act transparency requirements, whose core rules on identifying AI-generated content took effect on August 2, 2026.

Why It Matters for Brazilian Agencies and SMBs

For Brazilian agencies producing content with Claude for clients, the watermark does not change the delivered text or require any technical action, but it signals a trend likely to reach other AI tools used daily: traceability of AI-generated content is moving from a European regulatory exception to a market standard. It is worth factoring this into client contracts about content authorship, since detection tools based on this kind of mark could, in the future, be used by third parties, including Google in search results, to identify AI-generated text published as if it were original or entirely human-written.