SenseDefend

Synthetic-media instruments, by article and effective date

Detector, what it reports and what it cannot

A detector looks for a specific marking scheme and reports how confident it is. It answers one question about one scheme, and a negative answer is compatible with several very different states of the world. As of 2026-09-12.

What each result supports, and what it does notA found mark establishes that this scheme's mark is present. Nothing found is compatible with unmarked generated content, a stripped file, and a scheme the detector does not know.A mark was foundNo mark was foundWhat it establishesThis scheme's mark is presentNothing on its ownCompatible withMarked content, arriving intactUnmarked, stripped, or another schemeWhat would make it usefulAlready isKnowing the route preserves marksHow it is commonly readCorrectlyAs evidence of authenticityOne detector knows one scheme
Fig. 1 Useful for confirming origin and not for refuting it; used as if it could refute, it produces confident wrong answers.
Detector, as published. Recorded 2026-09-12.
What a detector reportsWhat that supports
A mark was foundThis scheme's mark is present
No mark was foundNothing, unless the route preserved marks
A score near the thresholdA judgement somebody has to make
An errorNothing about the content at all

Inclusion rule. Outputs of a detector for an embedded marking scheme, and the conclusions each supports. Order. From the informative result to the uninformative ones.

1One detector, one scheme

A detector has to know what it is looking for. Nothing obliges competing vendors to publish their schemes, so a tool built for one scheme finds nothing in material marked by another.

Which means cross-platform checking needs an agreement rather than a better detector. Until one exists, every checking route in practice is scoped to whoever built it.

2A negative result is the trap

Absence of a detected mark is regularly read as evidence that content is genuine. It is compatible with unmarked generated content, with a stripped file, and with a scheme the detector does not know.

So a detector is useful for confirming origin and not for refuting it. A tool used as if it could refute is worse than no tool, because it produces confident wrong answers.

3Thresholds trade two errors against each other

Set the threshold low and marks are reported in unmarked material. Set it high and marks in heavily processed material are missed. Both errors are real and neither can be eliminated by care.

Which of the two a system prefers is a policy choice, and it is almost never published. A detection rate without a false positive rate is half of a trade-off presented as a result.

4Who may run it is a separate fact

A detector can exist and be unavailable. Where a vendor describes its tooling as internal, the capability is real and reaches nobody holding a file without a relationship to that vendor.

A register keeps that on its own row for exactly this reason. What a mark is and who can read it are different questions, and only the second one matters to a stranger.

5Published material behind this entry

A term in this record: what it means, and what the published material says about it, each figure read on 2026-09-12. Related: Synthetic media, Open weights.