One caveat before the list: “watermarks text” does not mean “you can check for it.” Every deployed text watermark is keyed — only the provider holding the secret key can detect its own mark. No provider text-detection API is publicly available today. Our watermark explainer covers why.
Live adapter matrix
Provider states as reviewed on August 16, 2026 (served live by the detection service):
- Google announced SynthID-Text published by Google DeepMind (Nature 2024, open-sourced detector code), and deployed in Gemini — but no public verification API is available for third parties, so no adapter can go live.
- Anthropic announced Anthropic announced (Aug 11-14, 2026) that Claude models launched after Aug 2, 2026 watermark their text output; its detection API is announced but not yet public, so no adapter can go live.
- OpenAI unavailable OpenAI built a text watermarking method internally (reported 2024) but has not released it or any verification API; pending.
- Meta unavailable Meta published watermarking research (e.g. video/audio; Three Bricks for text) but ships no public text watermark verification; pending.
- Mistral unavailable No announced text watermarking or verification API from Mistral; pending.
- Microsoft unavailable Microsoft backs C2PA for media provenance but exposes no text watermark verification API; pending.
- xAI unavailable No announced text watermarking from xAI.
Provider status
Google — watermarking since 2024; no public detection API for text. Google has watermarked Gemini app and web text with SynthID-Text since 2024, and published the method in Nature [Google DeepMind; Dathathri et al., Nature, Oct 2024]. Detection remains in Google's hands: the SynthID Detector portal (announced May 20, 2025) is still waitlist-gated to journalists, media professionals, and researchers, and its text detection is not live; Google's enterprise Content Detection API (announced May 2026) currently accepts images only [Google blog, May 2025; DeepMind SynthID page; InfoQ, May 2026 — accessed Aug 16, 2026].
Anthropic — announced August 2026; detection API “coming soon.” Anthropic announced on August 11–14, 2026 that Claude models launched after August 2, 2026 watermark text globally, across all surfaces (API, Claude Code, AWS/GCP/Azure), using the SynthID-Text approach; older models follow “over the coming months.” A detection API that will “estimate the likelihood of Claude's involvement” is announced but has no docs, no waitlist, and no date [Anthropic, Aug 2026; TechCrunch, Aug 11, 2026].
OpenAI — pending; no text watermark shipped. OpenAI signed the EU Transparency Code of Practice (July 2026) and says its goal is “to expand provenance signals to all modalities including text,” but has shipped no text watermark. It built one internally around 2022–2024 and withheld it, citing false-positive and competitive concerns. Its images carry C2PA metadata and, since May 2026, SynthID [CityAM, Aug 12, 2026; InfoQ, May 2026; OpenAI help center — accessed Aug 16, 2026].
Meta — pending; signed the Code, nothing shipped for text. Meta signed the Code of Practice in July 2026; our searches as of August 16, 2026 found no shipped text watermark. Its published watermarking work (Stable Signature, video/audio marks in Meta AI) targets other media [EC signatory list, Aug 2026].
Mistral — pending; signed the Code, nothing shipped for text. Same position as Meta: a Code signatory with no text watermark found as of August 16, 2026 [EC signatory list, Aug 2026].
Microsoft — pending on text; metadata-first elsewhere. Microsoft signed the Code and already attaches C2PA Content Credentials to Copilot AI images, an audio watermark to Copilot audio overviews, and provenance metadata (which model, which app, when) to AI-generated content in Microsoft 365 — but no statistical text watermark [Microsoft Learn / Support, accessed Aug 16, 2026].
xAI — no commitment. xAI did not sign the Code of Practice — reported as the only major language-model maker to decline — and has announced no watermarking plans. It remains bound by the EU AI Act's marking obligation for output reaching the EU [EC signatory list; CityAM, Aug 12, 2026].
Why the “pending” column should shrink soon
Two EU deadlines force the issue. By December 2, 2026, generative systems already on the market must mark text output machine-readably (the AI Omnibus grace period ends). By February 2, 2027, Code signatories must provide interoperable detection — a public access point, an in-content signpost, or a consortium solution — so the outside world can actually check for their marks [ReedSmith; TechPolicy.Press; Stibbe — accessed Aug 16, 2026].
That February 2027 deadline is our promise to you: the day a provider opens a detection route, we connect to it and this page changes. Check back — or just run a scan; the technical details on every scan result state which provider watermark checks are live, from the same live status feed as the matrix above.
Sources
- Google DeepMind, SynthID page, deepmind.google/models/synthid/ (accessed Aug 16, 2026); Dathathri et al., Nature (Oct 2024); Google blog, SynthID Detector announcement (May 20, 2025); InfoQ, AI Content Detection API (May 2026)
- Anthropic, Claude text watermark announcement (Aug 2026); TechCrunch (Aug 11 and 15, 2026)
- European Commission, Code of Practice signatory list (updated Aug 12, 2026)
- CityAM (Aug 12, 2026); OpenAI help center art. 8912793 (accessed Aug 16, 2026)
- Microsoft Learn, Copilot watermarks documentation (accessed Aug 16, 2026)
- ReedSmith; TechPolicy.Press; Stibbe — analyses of the Code's detection and interoperability deadlines (accessed Aug 16, 2026)