AI Watermark Status: Who Marks What, Today
Which AI tools currently watermark content, how those marks work, and whether they can be detected or removed.
As of August 30, 2026, 7 provider surfaces publish a machine-readable mark on their output. This page tracks which providers watermark what, whether a detector exists, and when each verdict was last checked.
| PROVIDER | CONTENT | WATERMARK | DETECTION | REMOVAL | CHECKED |
|---|---|---|---|---|---|
| Claude Anthropic | Text | Statistical text watermark | Announced: Announced Official detector not yet public | Unknown: Not local Not locally removable | Aug 30 |
| Claude Anthropic | Generated files | C2PA metadata | Clean: Supported Content Credentials readable locally | Clean: Strippable Metadata only; wording unaffected | Aug 30 |
| ChatGPT OpenAI | Text | None confirmed | Unsupported: None No confirmed statistical detector | Clean: Partial Hidden Unicode removable where present | Aug 30 |
| GPT image OpenAI | Image | Pixel watermark + C2PA metadata | Unknown: Limited C2PA readable locally; pixels are not | Clean: Strippable Metadata only; pixels unverifiable | Aug 30 |
| ChatGPT audio OpenAI | Audio | Embedded audio watermark | Unknown: Limited Verifiable through OpenAI’s own tool | Unknown: Not local Embedded in the signal | Aug 30 |
| Gemini Google | Text | Statistical text watermark | Unknown: Limited Detector portal is waitlist-only | Unknown: Not local Rewrite only; not a character strip | Aug 30 |
| Gemini Google | Image | Pixel watermark + C2PA metadata | Unknown: Limited Pixel checks live in Google Search | Clean: Strippable Metadata strips; pixels stay | Aug 30 |
| Firefly Adobe | Image | Content Credentials (C2PA) | Clean: Supported C2PA manifest readable locally | Clean: Strippable Manifest removable; provenance breaks | Aug 30 |
| Any image file EXIF · XMP · PNG text chunks | Image | File metadata | Clean: Supported Parsed locally in the browser | Clean: Strippable Removed and verified by rescan | Aug 30 |
| Other text models Copilot · Grok · Meta AI · Perplexity · Mistral · DeepSeek | Text | None announced | Unsupported: None No text watermark announced | Clean: Partial Hidden Unicode removable where present | Aug 30 |
States: ✓ supported · announced · ? limited / unknown. Each provider name links to its page.
Cite this page- As of August 30, 2026, no public Claude watermark detector exists. Anthropic said a detection API is coming and has not given a date.
- As of August 30, 2026, 7 Claude models are shipped and selectable. All of them predate Anthropic's launch-date cutoff of August 2, 2026. Status: Unknown.
- Google's SynthID Detector portal is waitlist-only for journalists, media professionals, and researchers as of August 30, 2026.
- OpenAI has not deployed a text watermark for ChatGPT or the API as of August 30, 2026. Images from ChatGPT carry SynthID in the pixels and C2PA Content Credentials in the file.
- The EU AI Act Article 50 deadline for retrofitting pre-existing systems is December 2, 2026.
DetectMark, "AI Watermark Status", last checked August 30, 2026. https://detectmark.com/ai-watermark-status
The underlying data is published under CC BY 4.0. Reuse it with attribution.
About this tracker
It is reviewed weekly, and out of band whenever a provider ships a change or a detector opens.
Anthropic (Claude), OpenAI (ChatGPT and GPT image), Google (Gemini and SynthID), Adobe (Firefly and Content Credentials), and the text models that have announced nothing: Copilot, Grok, Meta AI, Perplexity, Mistral and DeepSeek.
The labels are Clean, Found, Unknown, Unsupported, Announced, Supported, Limited, Partial and None. Clean means we ran the check and found nothing. Found means we ran the check and a known signal is present. Unknown means we cannot verify, usually because the provider has not opened a detector. Announced means the provider committed to a mark but nothing verifiable ships yet. Supported means the check can actually be performed, in DetectMark or in a public tool. Limited means a check exists but access is restricted, such as a waitlist portal or a check that only runs inside the provider's own product. Partial means only some of what is present can be checked or removed, such as hidden Unicode in a paste that may also carry a statistical mark. None means the provider has not announced any mark for this content type. Unsupported means the check does not apply. Never a percentage, and never an AI likelihood score.
Yes. Each provider page carries the source list behind its row, and the machine-readable export includes a source URL per record.
It is the date the verdict was last verified against the provider's own documentation, an independent test, or a published policy. It is never a generated or estimated date.