Insights

ONVIF signs video at the camera. Can it survive deepfakes?

On 26 August 2026, ONVIF announced the Release Candidate for its Media Signing Add-on. The spec gives IP cameras a standardised way to cryptographically sign video at the moment it is encoded. The goal is straightforward: allow anyone downstream to verify whether footage originated from a specific device and whether it has been altered since.

The timing is not accidental. AI-generated and manipulated video has reached the point where a courtroom opponent or an insurance adjuster can plausibly demand proof that surveillance footage is unmodified. ONVIF Board Chairman Leo Levit described the problem directly at the announcement: proving authenticity is central to trust in video, and the industry needs a shared method to defend it.

How it works

The mechanism is technical and deliberately low-bandwidth. The camera hashes each video frame using SHA-256. At intervals, the device generates a cryptographic signature over the accumulated hashes using a private key stored on the camera. That signature data travels inside the video stream itself as a Supplemental Enhancement Information (SEI) NAL Unit, specifically a “user data unregistered” type with a fixed UUID (005bc93f-2d71-5e95-ada4-796f90877a6f). The SEI data does not affect video decoding or playback. Players that do not recognise it simply ignore it.

A conformant verification tool reads the SEI data, reconstructs the hash chain, and checks the signatures against the device’s public certificate. Any edit after encoding, a re-encode or a crop, breaks the hash chain and verification fails. The tool can also link the video to the specific camera that produced it, providing both integrity and provenance.

ONVIF released the implementation as open-source code on GitHub (onvif/media-signing-framework). The library in C handles both signing and verification. The build chain requires OpenSSL 3.0 or later, meson 0.49+, and optionally GStreamer for the example applications. The signing logic is abstracted into a plugin interface because key management is device-specific: how a camera accesses its private key is a hardware question that each manufacturer must answer.

The current release covers video only. Audio signing is planned for a future version.

Why this matters for evidence

Australian courts have been grappling with digital evidence authenticity for years, but the rise of generative video has sharpened the problem. The Evidence Act 1995 (Cth) and state equivalents require that a party tendering electronic evidence can demonstrate its integrity. For surveillance footage, that usually means a chain-of-custody affidavit from whoever exported the recording. The affidavit says the footage has not been altered. It does not prove it cryptographically.

Media Signing changes the verification model from human attestation to mathematical proof. If a camera conforming to the add-on produces signed footage, and that footage verifies on a conformant tool, the footage has provably not been modified since the camera encoded it. That is a different standard to “I exported this from the VMS and nobody touched it.”

The catch is adoption. The specification is a Release Candidate. ONVIF expects the final standard to ship by the end of 2026. Camera manufacturers must then implement the signing interface in their firmware, and VMS vendors must build verification into their export and playback tools. ONVIF conformance requires a product to also conform to at least one ONVIF profile, so Media Signing sits on top of existing interoperability rather than replacing it.

For Australian integrators and system specifiers, this represents a procurement signal. A security tender that specifies ONVIF Media Signing conformance is asking for footage that can defend itself in a legal setting. Sites with evidential requirements - law enforcement, corrections, critical infrastructure, corporate investigations - have a reason to track this standard now rather than after it ships.

The framework’s plugin architecture also matters. The private key handling question is not trivial. If the key is stored in software, extraction compromises the entire trust model. Camera vendors implementing signing with hardware-backed key storage (a TPM or secure enclave on the SoC) will produce footage with a stronger assurance claim. Specifiers should be asking about key storage, not just conformance.

What to watch

Deployment depends on vendor firmware and on VMS integration, and neither is inside the specification’s control. ONVIF’s 500+ member list includes every major manufacturer, but firmware roadmaps are not public commitments, and the vendors who ship signing support in their next generation of firmware will set the pace. The same problem applies to the tools operators already use: if checking a signature needs a standalone command-line utility, adoption stalls at the integrator level, so verification has to land inside VMS export and playback workflows.

Australian courts are the other variable. No reported decision has yet addressed ONVIF Media Signing specifically. When the first case presents signed footage and opposing counsel cannot demonstrate tampering, the precedent will be worth more than any specification document.

Sources

  • ONVIF press release, “ONVIF to Strengthen the Authenticity of Video Surveillance Footage with Media Signing Add-on,” 26 August 2026
  • ONVIF Media Signing Add-on specification (Release Candidate) and FAQ, onvif.org
  • ONVIF Media Signing Framework source code, github.com/onvif/media-signing-framework
  • Security.World, “ONVIF Advances Video Authenticity With Media Signing,” 13 September 2026
  • GSX 2026 educational programming, “Can Security Video Survive Generative AI?” Roberto Licari, ONVIF Ambassador, 14 September 2026
  • Evidence Act 1995 (Cth), Part 4.6, Authentication and identification