Anthropic Claude Watermark Checker

Anthropic Claude Watermark Checker takes images, video, or audio and returns structured JSON on whether the file carries a Claude Content Credential. That is Anthropic’s public, machine-readable file mark: signed C2PA provenance metadata. The Actor answers the same question as Anthropic’s official checker at claude.com/check-files: does this file show a Claude-issued credential? The headline output field is isAiGenerated when a trusted Claude credential is present.

Claude also watermarks generated text, but Anthropic’s text Detection API is still private preview. This Actor does not guess from writing style and does not accept pasted essays.

---

What is Anthropic Claude Watermark Checker?

Upload a still, clip, or sound file Claude may have exported (a chart PNG, an SVG, a JPEG, a short MP4). The Actor reads the embedded Content Credential locally and verifies the signature against Anthropic’s Content Credentials Root CA — the same trust anchor Anthropic’s in-browser checker uses. You get one dataset row per file: watermarkDetected, matchesTargetProvider, C2PA validation state, and a limitations note.

No Claude login. No Gemini. The file is not sent to Anthropic; verification runs in the Actor.

---

What can this Actor do?

• Check images (JPG, PNG, GIF, WEBP, TIFF, HEIC, AVIF, SVG, DNG, JXL)
• Check video (MP4, MOV, AVI) and audio (WAV, MP3, M4A, FLAC)
• Return isAiGenerated when a trusted Claude Content Credential is present
• Match Anthropic’s public file checker: credential found only if the C2PA manifest reads and the signature is Trusted under Anthropic’s CA
• Charge per file (local cryptographic check — no LLM tokens)
• Export JSON, CSV, or Excel, or pull results via the Apify API

It does not decode Claude’s embedded text watermark. That detector is not public.

---

Why check Claude files for Content Credentials?

From August 2026, supported Claude models attach signed C2PA metadata to generated files (for example .png, .jpg, .svg). Teams that already hold the export need a repeatable JSON row, not a screenshot of a browser drop zone.

Typical jobs:

1. Newsroom provenance — log whether a graphic was processed with Claude
2. Brand safety — screen UGC stills and clips for Claude credentials
3. Marketplace QA — catch Claude-exported product charts
4. Trust and safety — label files that carry a Claude credential
5. Legal extras — attach a C2PA result next to produced media
6. Dataset labeling — tag training images that still have credentials
7. Publisher intake — check freelance illustrations and motion
8. Education tools — show a credential flag, not a red card
9. Ad review — screen generated creatives
10. Internal audits — sample Drive/Slack media dumps
11. Policy logging — EU AI Act-style machine-readable file marks
12. Research studies — batch-read C2PA on Claude-era exports

---

How to use Anthropic Claude Watermark Checker

1. Open the Actor in Apify Console
2. Upload Files (images, video, or audio — up to 100 MB each)
3. Click Start
4. Download the dataset or read it via the Apify API

Input example

{
  "files": ["https://example.com/claude-chart.png"],
  "maxItems": 5
}

Field	Type	Default	Description
files	string[]	—	File URLs to check
maxItems	integer	5	Cap on files this run

---

Output

Each dataset item is one file. watermarkDetected is true only when the Content Credential verifies as Trusted against Anthropic’s CA (the official checker’s rule). model is the detector family (c2pa).

{
  "id": "a1b2c3d4e5f67890",
  "sourceType": "image",
  "sourceName": "claude-chart.png",
  "sourceUrl": "https://example.com/claude-chart.png",
  "imageUrl": "https://example.com/claude-chart.png",
  "mediaKind": "image",
  "mediaType": "image/png",
  "fileSizeBytes": 84211,
  "isAiGenerated": true,
  "watermarkDetected": true,
  "watermarkType": "c2pa",
  "matchesTargetProvider": true,
  "likelySource": "claude",
  "likelihood": "high",
  "confidence": 1,
  "detectionMethod": "c2pa",
  "rationale": "Signed Claude Content Credential verified against Anthropic's Content Credentials Root CA.",
  "limitations": "A Claude Content Credential means the file was produced or processed with Claude. It is not proof of original authorship.",
  "c2paValidationState": "Trusted",
  "c2paClaimGenerator": "Claude",
  "model": "c2pa",
  "targetProvider": "anthropic-claude",
  "error": null,
  "checkedAt": "2026-09-03T12:00:00.000Z"
}

---

Pricing

The check is local C2PA verification. Cost does not scale with tokens.

Event	When	Price (FREE)
text-checked	Once per successful file check	$0.005
Actor start	Per run (× memory GB)	platform default

Failed downloads are not billed.

---

What this does and does not prove

Anthropic’s own docs are the source of truth:

• A detected credential signals the file may have been processed with Claude. It does not identify the user, and Claude may only have edited or converted existing material.
• No credential does not mean the file is human-made. Edits, conversion, compression, screenshots, older models, and unsupported surfaces can strip or omit the mark.
• Text is a different mark. Use Anthropic’s Detection API when you have access; do not paste essays into this Actor.

For a Google SynthID-Text check on writing, use Google Gemini Watermark Checker. For ChatGPT images and audio, use OpenAI ChatGPT Watermark Checker.

---

Is it legal to check Content Credentials?

This Actor only analyzes files you provide. It does not log into Claude. Do not treat a credential as legal proof of authorship.

---

Tips

• Prefer original exports from Claude, not screenshots
• Files over 100 MB are rejected (same cap as Anthropic’s checker)
• Start with maxItems 1–3 while you learn the output shape
• Cross-check a sample in Claude’s file checker if you need a second opinion on the same file

---

Image Credit

Image credit: Anthropic