Verifying Claude-Generated Files

This utility allows users to determine if a specific file was produced using Claude.

🛠️ How to Use the Tool

• Drag and drop your file into the upload area.
• Or, click to browse and select a file.

Note: By utilizing this service, you are agreeing to the terms outlined in Anthropic’s Usage Policy and acknowledging their Privacy Policy.

---

📁 Technical Specifications

The tool operates entirely within your browser environment. The maximum allowable file size is $100\text{ MB}$.

Supported File Extensions

The following formats are compatible with the checker:

---

🔍 Understanding the Detection Process

Claude employs two distinct methods for marking its output:

1. Textual Watermarking: Integrated directly into the prose of essays or emails.
2. Content Credentials: Signed provenance metadata embedded in downloadable files (e.g., a generated chart).

Crucially, this specific tool only detects the content credential.

It cannot analyze the actual content of the file nor can it reveal any private information regarding the user who generated the file.

The C2PA Standard

When Claude creates a supported file, it inserts a cryptographically signed note into the metadata. This follows the C2PA (Coalition for Content Provenance and Authenticity) open industry standard.

Technical Implementation

The metadata acts like a digital seal. Conceptually, the check looks for a signature similar to this:

{
  "provenance": {
    "tool": "Claude",
    "signature": "0xabc123...cryptographic_hash",
    "standard": "C2PA"
  }
}

---

🌐 Ecosystem and Alternatives

Because C2PA is a universal standard, any tool compatible with C2PA can read these credentials. Other AI developers have implemented similar verification systems, including:

• OpenAI
• Google DeepMind
• Gemini

What about text?

To identify watermarks within text (rather than files), Anthropic provides a Detection API. This is currently limited to a private preview for specific organizations to satisfy requirements under $\text{EU Law}$.

---

🔒 Privacy and Data Handling

Your privacy is maintained through the following architecture:

• Local Processing: Your file never leaves your device.
• Selective Reading: The tool is designed to read only the embedded credential $\rightarrow$ it does not read the actual content of the file.