Detection you can defend.

Sighting.ai was born from a simple observation: as AI writing tools got better, detection tools answered with bare percentages and scary dials — verdicts that fall apart the moment someone asks “how do you know?”

We build the opposite. Every detection returns what the engine actually found: for text, the exact passages that look AI-written and the signals that flagged them; for images and video, the provenance we could and could not establish. Always with a confidence range that admits what the engine doesn't know.

That cuts both ways, and the second half is the harder one. A verdict of “human” is only ever issued on positive provenance of the exact bytes in front of us — a signed capture manifest from a signer we accept — because a score low enough to look innocent is also what a screenshotted AI image looks like. Images additionally get a pixel classifier, which can identify an AI picture and can never clear one. Video gets no classifier at all. A signed file can be settled outright; an unsigned one gets a real provenance report and an uncertain verdict, because nothing that can honestly decide it exists yet.

The same standard is why we do not detect AI-written source code. We measure the file and show the measurements, but we return no verdict about its author, because no method we can run tells AI-written code from human-written code reliably — and the way those methods fail is by flagging human developers. Benchmarks find them misclassifying human code as AI more often than they miss AI code. Shipping that would mean accusing people who did their own work, so we do not ship it.

We serve educators, editors, and developers who need verdicts they can hand to a third party — and have them survive scrutiny.