Check a file. Then check the facts.

Every check runs three independent lines of evidence: a trained image-detection model on your device, a structural read of the file's format, metadata, and edit traces, and local validation of C2PA Content Credentials with the official Content Authenticity Initiative SDK. Text checks measure writing patterns common in generated prose. We work on detection every day — as new AI generators appear, our models and trust data are updated to keep pace. And when the evidence is weak, AICheck says so instead of guessing: no detector can prove who made something, so each clue is shown separately and honestly. Treat any score as a strong lead to verify against the original source, not a final verdict.

Yes, with honest limits. A video first checks supported embedded Content Credentials locally, then requests 8 evenly spaced still frames through the image model and shows the median and highest successful scores. A timed-out or failed seek is labeled and not counted as a sampled frame. It is not a full-video analysis: only the first ten minutes are sampled, and audio and motion over time are not analyzed. A screen check watches the screen or window you choose for about four seconds, samples up to 8 frames from it (the result states how many were actually scored), runs each through the same image model, and reports the median and highest scores — then stops sharing. No recording is written to a file; only transient frames exist, and they stay on-device. On iPhone, iPad, and in the mobile apps, screen capture is not available, so AICheck gives screenshot instructions instead. Frames stay on-device.

Not by default. Text and image checks run on your device and the local result appears first. On the web, an image up to 3 MiB is uploaded only after an active AICheck Pro subscription enables the cloud box and you tick it for that check. The configured inference provider receives that image. AICheck keeps an AES-GCM-encrypted temporary queue payload accessible to the job for up to 15 minutes, deletes it promptly on completion or failure, and removes expired records through bounded cleanup; account deletion purges them immediately. QStash receives only an opaque job ID plus orchestration metadata. Native app scans remain on-device in this release. See Privacy for details.

Content Credentials (C2PA) can carry signed provenance claims about how media was created or changed. For supported embedded credentials, AICheck locally checks the manifest, signature, connection to the file, and signer trust using the official Content Authenticity Initiative SDK and a bundled checksum-pinned official trust-list snapshot. It does not fetch remote manifests. If the validator, WebAssembly runtime, or trust list cannot complete the check, AICheck reports unavailable rather than calling the credential invalid. A valid credential records signed provenance; it does not make the content true, and a missing credential does not make it untrustworthy.

Web AICheck Pro subscriptions are billed by Stripe. While Pro is active, choose Manage Pro at the top of this page to open Stripe's secure portal and update or cancel it. Your Stripe receipt and support remain available as fallbacks. Native AICheck Unlimited subscriptions are managed in your App Store or Google Play account settings.