SenseDefend

Synthetic-media instruments, by article and effective date

ElevenLabs on the marks its own detector can and cannot find

The page limits its own detector explicitly: it recognises marks written by this vendor and not those of anybody else. That sentence is more useful to a reader than the broader one the vendor could have written. As of 2026-09-22.

What a negative result from a single-vendor detector meansGiven a file known to come from a platform, asking that platform whether it made it is nearly redundant. The interesting case is a clip of unknown origin, where a negative answer means only that this vendor's mark is absent.The detector reports no mark. What follows?The file came from this vendorSomething is oddEither the marking was out of scope, or theroute stripped itOrigin unknownNothing followsAnother platform's mark would not be foundby this detector at allThe page states the limit itself, which is why it is recorded
Fig. 1 A reader who reads absence as authenticity has been misled by the tool, not by the page.
ElevenLabs on the reading route column, and what that column leaves open. Recorded 2026-09-22.
PointWhat the record holds
Reading routeThe vendor's own detector, for marks it embedded itself
Read fromProduct documentation for the checking tool
What it does not settleWhat to do with audio whose origin is unknown

Inclusion rule. One cell of this generator's entry, quoted from the page the entry names. Coverage, commentary and third-hand summaries are not admitted, in either direction. Order. Fixed order: the cell, the page it came off, then the limit on reading it.

1A checking tool scoped to one vendor answers the easy question

Given a file known to come from a platform, asking that platform whether it made it is nearly redundant. The interesting case is a clip of unknown origin, and a single-vendor detector returns nothing useful about it.

A negative result from such a tool means this vendor's mark is absent. It does not mean the audio is authentic, and a reader who treats the two as the same thing has been misled by the tool rather than by the page.

2The same page narrows the coverage a second time

Alongside the detector's reach, the documentation states that models from other providers offered through the platform are generally not watermarked unless stated otherwise. So the marking is scoped inside the product too.

Two narrowings on one page is unusual, and both point the same way: the vendor is describing something that works inside its own systems. The record keeps the reach and the coverage on separate rows.

3Why a per-vendor detector is not a failure of design

Audio watermarking schemes are not interchangeable. A detector has to know what it is looking for, and nothing obliges competitors to publish their schemes, so cross-platform checking needs an agreement rather than a better detector.

That is the gap the codes-of-practice paragraph anticipates for this whole subject. Until something fills it, a per-vendor route is the only route, and saying so is the most accurate thing a page can do.

  • ElevenLabs
    The detector can only find watermarks embedded by this vendor and will not find those of other platformsthe reach of the checking routeElevenLabs, documentation / recorded 2026-09-22
  • Article 50(2)
    Technical solutions must be effective, interoperable, robust and reliable as far as technically feasiblequalifier on the marking dutyAI Act Explorer, Article 50 / recorded 2026-09-12

4Sources

Wording taken from the documentation at elevenlabs.io, consulted 2026-09-22. All four rows for this generator sit on ElevenLabs; the same column across every entry is read down on Reading route. Filed beside it: Nova Reel on public checking tools, Designer on the credentials site.