All Radar picks

X OPPORTUNITY RADAR / THE BRIEF

The Truth Machine Has a Guest List.

Google opens its watermark detector worldwide in English. Useful provenance, yes. Universal truth machine, no—and the official pages disagree on Apple’s timing.

Information Overload
2 min readSources
Conceptual editorial illustration of an industrial inspection kiosk scanning a framed image, a video strip and an audio waveform for invisible watermarks, while similar unmarked media slips around the machine through a shadowed side lane.
AI-generated conceptual editorial illustration. It is not documentary evidence or a screenshot of Google’s detector.

The bouncer can read the wristband. That does not make him an oracle, droogies.

Google’s October 29 update to its October 7 announcement says SynthID Detector is now available worldwide in English. People can submit images, video and audio to check for participating AI watermarks. That is a useful public doorway into provenance—not a machine that decides whether the internet is telling the truth.[1]

What gets past the velvet rope

DeepMind describes SynthID as a system that embeds imperceptible marks in generated media, then checks for those marks. The public detector follows testing with journalists and media professionals. Its remit is specific: recognize this watermark system, not every synthetic object wearing a respectable coat.[2]

Google reports more than 180 billion watermarked images and videos, plus 240,000 years of audio. It also says verification features in Search, Gemini and Chrome handle over a million daily requests. Those are the operator’s figures, not an independent audit of detection accuracy or universal coverage.[1]

The Apple-shaped footnote

The announcement lists OpenAI, NVIDIA, Kakao and Apple among the partners. DeepMind’s product page lists the same companies but describes Apple as coming soon. The official pages disagree on timing. Current Apple support is therefore not established here; no triumphant rollout graphic for that one, my little marketing malchicks.[1][2]

A watermark is not a verdict

Here is the editorial distinction worth keeping: detecting a participating mark can provide evidence about how media was made. It cannot establish that the depicted claim is accurate or that the file is being presented honestly. Equally, failing to find that mark cannot certify human authorship.[2]

A real photograph with a false caption can still do its dirty work without any generative model involved. Provenance and truth are different questions. This is a helpful tool for the first; promoting it to judge and jury for the second would be the old confidence trick with a shinier scanner.

FOLLOW THE SOURCES.

Checked Oct 29, 2026. Numbered references point to the evidence behind factual passages. Commentary is our own.

  1. We’re making it easier to identify AI-generated content globally. (opens in a new tab)GoogleBack to cited passage
  2. Google SynthID (opens in a new tab)Google DeepMindBack to cited passage
Artificial intelligenceMedia provenanceWatermarkingGoogle
Back to the latest picks