Anthropic has released Claude Content Checker, a free tool that tells you whether Claude was involved in creating or editing an image, video, or audio file simply by dragging it onto a web page. The check runs entirely inside the browser, so the file is never sent to Anthropic's servers, and no account is required. Under the hood, the tool reads the C2PA content credential embedded in files Claude generates. Notably, Anthropic is just as careful to explain what the tool cannot tell you as what it can.

Drag, drop, and the file stays with you

The tool lives on the check-content page at claude.com. To use it, you drop the file you want to inspect onto the page or click to select it. Supported image formats are JPG, PNG, GIF, WEBP, TIFF, HEIC, AVIF, SVG, DNG, and JXL; video formats are MP4, MOV, and AVI; and audio formats are WAV, MP3, M4A, and FLAC. Each file can be up to 100 MB.

Anthropic stresses that all processing happens on the user's device. The checker reads only the signed record embedded in the file, and the file itself is never stored or reused. That makes it practical for checking confidential documents or unreleased material without uploading anything externally.

What it reads: a signed note in C2PA format

The verdict is based on metadata that follows the C2PA (Coalition for Content Provenance and Authenticity) industry standard. Camera makers and photo-editing software use the same mechanism to record which device captured an image and which software edited it. When Claude generates a file in a supported format, it automatically attaches cryptographically signed provenance information. Download a chart Claude made, for example, and the image carries a record showing that Claude processed it.

The checker looks for that record and, if it finds one pointing to Claude, reports that traces of Claude's involvement are present. This is strictly a metadata match. It is a different principle from so-called AI detectors that analyze the look of an image or the wording of a text to estimate how "AI-like" it is.

"Not found" does not mean "not used"

Anthropic is explicit about what the tool cannot do.

First, the result only indicates whether the file was processed by Claude; it says nothing about the accuracy or value of the content. Because people use Claude not just to generate files but also to convert and revise them, a record can be present even when the ideas or underlying data came from somewhere else.

Second, the absence of a record does not prove the file was not made with Claude. Format conversion, re-saving, or taking a screenshot can strip the metadata, and the trace disappears with it. Files produced by models, features, or platforms that do not support provenance marking never had a record in the first place.

Third, the record contains no information that identifies the user. The tool cannot tell you who made a file, only whether Claude was involved.

Part of EU AI Act compliance, separate from text watermarking

The tool is one piece of the AI-content marking program Anthropic has been rolling out since August. The company has signed the transparency code of practice under Article 50(2) of the EU AI Act, and Claude models launched on or after August 2, 2026 are designed to embed watermarks in generated text and attach signed provenance metadata to generated files. The models currently supported are Fable 5.1 and Mythos 5.1, with support for earlier models to follow.

Marking applies across the Claude API, the Claude app, Claude Code, Claude Cowork, and Claude Tag, as well as access through AWS, Google Cloud, and Microsoft Foundry, in every region where Claude is offered.

Detection of the text watermark, however, is still in private preview. It is available only to organizations that need it for verification under EU law, such as regulators, media outlets, fact-checking groups, and researchers, and enterprises using it for their own compliance must apply for access. This file checker, which reads content credentials, is the first publicly available tool anyone can use. We covered how the text watermark works in a separate article in August.

Summary

Claude Content Checker is a free, browser-based tool that uses C2PA provenance data to confirm whether a file was processed by Claude. Files never leave your device, no sign-up is needed, and the supported formats span images, video, and audio. On the other hand, it can only find records that are still there: files whose metadata has been stripped, or that were made by unsupported models, cannot be judged. The right way to think of it is not as a tool that detects AI content, but as a tool that checks whether the mark Claude left behind is still intact.