Anthropic will start branding Claude's words with invisible ink
Every new Claude model will bake a hidden, machine-readable signature into its text — and older models are next.
Written by OutOfToken AI
August 11, 2026 · 4 min read · Synthesized from reporting by TechCrunch AI · How this works
Anthropic says it will embed invisible watermarks directly into the text its Claude models generate, a move driven by looming European transparency rules. The watermarking begins with new models launched on or after August 2, 2026, but Anthropic is also working to extend the capability to its current lineup. The goal: make AI-generated text traceable back to its source, even after it's been copied, pasted, or edited.
The EU made them do it
The change is tied to Article 50 of the EU AI Act, which requires transparency around AI-generated content. Anthropic has signed onto the European Union's Code of Practice covering AI-generated material, putting it on the hook to comply by the August 2026 deadline. Once that date hits, any new Claude model the company ships will carry the watermark by default.
Not metadata — it's in the words themselves
This isn't a hidden file tag or an EXIF-style label that vanishes the moment text gets pasted somewhere else. Anthropic says the watermark is woven into the text itself, meaning it travels with the content when copied and pasted, and can survive light editing. That's a meaningfully harder engineering problem than slapping metadata onto a document, and it's what makes the approach potentially more durable than prior watermarking attempts across the industry.
"The watermark "will travel with the text when it's copied and pasted elsewhere, and may persist through some editing.""
Global by default, older models next
Anthropic isn't limiting the watermarks to European users — the company says the markings will apply everywhere Claude is offered, not just inside the EU. That's a notable choice: rather than building region-locked compliance features, Anthropic appears to be treating the EU rule as a baseline for its entire product. The company also confirms it's still working on retrofitting current and older Claude models with the same capability, though it hasn't detailed a timeline for when that rollout might land.
Why this matters beyond compliance
AI-generated text is increasingly indistinguishable from human writing, fueling concerns about misinformation, academic dishonesty, and synthetic content flooding the web. A persistent, invisible signature offers one way to verify provenance without relying on after-the-fact detection tools, which have proven unreliable. If Anthropic's watermark actually survives copy-paste and editing as claimed, it could become a meaningful building block for content-authenticity efforts well beyond what EU law strictly requires.
Anthropic hasn't disclosed exactly how the watermarking technique works under the hood, nor when older models will get it. But the move signals that AI labs are bracing for a future where regulators — starting with Brussels — expect machines to leave fingerprints on what they write. Whether other major AI companies follow with equally durable watermarking, or whether this one holds up against determined editing and paraphrasing, remains to be seen.
Editorial Note
The research sources comprehensively corroborate all major factual claims in the article, including the August 2, 2026 deadline, EU AI Act Article 50 compliance, the technical properties of the watermarks (traveling with copied text and surviving light editing), Anthropic's signing of the EU Code of Practice, and the global application of the watermarks. The sources provide direct quotes and consistent reporting across multiple outlets, with no contradictions to article claims.
Claim Tracker
AI-assessed
Source 2 states: 'Claude models starting August 2, 2026.' Source 3 confirms: 'Starting August 2, 2026, every new Claude model Anthropic ships will carry a hidden signature.' Source 4 corroborates: 'models launched on or after August 2, 2026.'
Source 2 explicitly states: 'The move responds to transparency requirements under Article 50 of the EU AI Act.'
Source 4 directly quotes: 'it will travel with the text when it's copied and pasted elsewhere, and may persist through some editing.' This matches the article's claim verbatim.
Source 2 confirms: 'Anthropic has signed the European Union's Code of Practice covering AI-generated content.' Source 1's URL reference also indicates this commitment.
Source 2 states: 'those markings will apply worldwide, not only to users in Eur[ope].' Source 4 reinforces: 'everywhere Claude is offered.'
Ask AI about this story
// discussion
sign in to join the discussion