How US-based Truepic and UK-based Serelay discern true images from fakes by using proprietary algorithms to automatically verify photos the moment they're made
Karen Hao / MIT Technology Review :
Context & Ripple Effects
This 2018 profile lands between two signals that defined the image-authentication problem: a February [[a:926892|analysis showing that Facebook's and Google's own detection algorithms often fail on doctored media]], and Truepic's June $8M Series A months later. The shared thesis of Truepic and Serelay is that verification belongs at the moment of capture, not in post-hoc forensics that detectors keep losing.
The piece reads differently now than it did then: Truepic went on to raise a $26M Series B led by Microsoft's M12 and co-launched the C2PA provenance standard with Adobe, Arm, Intel, and Microsoft, while Nikon, Sony, and Canon began building digital-signature capture tech — the capture-time approach this article documented becoming industry infrastructure.
First-order effects
- Platforms relying on third-party verification — Reddit used Truepic to authenticate identities in AMAs — get a check that runs when the photo is taken, closing the gap left by upload-time detectors that the WSJ analysis showed struggling.
Second-order effects
- Microsoft's M12 backing of Truepic foreshadowed the C2PA standard co-founded by Adobe, Arm, Intel, Microsoft, and Truepic, pulling provenance from startup feature to cross-industry specification.
Third-order effects
- If capture-time attestation holds, camera hardware becomes the trust boundary: Nikon, Sony, and Canon embedding digital signatures at the sensor shifts authentication from software guessing after the fact to cryptography baked into the device.
The trend: Image authenticity is migrating from post-hoc fake detection toward cryptographic provenance established at the moment of capture, with startups, platform giants, and camera makers converging on the same layer.