AI-generated videos have become so good that it’s hard to tell fact from fiction. And in many cases, by the time a video is fact-checked and proven to have been generated by AI, those who watched it may continue to believe that what they saw was true.
This big trust issue could soon be resolved with ASX-listed Ion Video (code IOV) filing a patent with IP Australia to determine whether a video is real or AI-generated.
The way this technology works is that instead of analyzing pixels on the screen or relying on metadata that can be removed, it authenticates the video at a binary sample level and creates a cryptographic record.
When detection between human and synthetic content is required, Ion’s technology infrastructure not only allows the content to be accepted, but passes it through what Ion calls the Authentication Policy Engine. This engine unifies various schemes such as C2PA content credentials, camera sensor fingerprints, codec forensics, AI detectors, studio authentication, and capture device registries under a single signed policy-gated encryption record.
Finbar O’Hanlon, Ion’s head of innovation, said that when a video is assembled and played through the company’s virtual video interface, “it’s a great experience.
If desired, the technology can verify each binary sample against a registered hash in real time. “If the bytes match, the sample is real. If they don’t match, the sample is flagged as unverified, modified, or potentially synthetic.
“Our goal with this patent is to provide a way to prove that a video is genuine. Our technology has the ability to verify content at the binary sample level using a variety of methods, including cryptographic hashes, so the proof is reflected in the bytes themselves rather than pixels on a screen. It is designed to remain effective as generative AI advances and withstand daily changes to files such as re-encoding and compression.”
He said the trusts that support video, including news, entertainment, financial services, government and corporate communications, are under increasing pressure.
“Existing approaches to this problem are largely reactive. Detection tools that analyze the decoded pixels of a video are in constant competition with generative models, and each new detector has the potential to be outcompeted by the next generation of synthesis.”
“Ovenance metadata, such as content credentials, can be removed or lost during re-encoding and redistribution. Invisible watermarks are vulnerable to common transformations such as compression and cropping. Additionally, hardware capture signatures only verify a single moment of capture. None of these methods together constitute multiple trust signals, and none operates at the binary sample level.”
He said Aeon’s technology takes a different approach. Rather than treating the video as a sealed, finished file and attaching validation externally, we bind the authentication directly to the underlying structure of the internal video DNA (the file’s constituent parts) and record the authentication in an immutable registry that exists independently of the file.
He said studios, broadcasters, government agencies and businesses can each create their own policies and define which trusted authentication sources content must meet before being entered into the trust system.
O’Hanlon said the patent filing completes the four-layer architecture of Aeon’s virtual video portfolio. This forms an end-to-end stack across four functions: virtualization, recording, management, and authentication, in line with Ion’s three existing patents.
The core patent and its accompanying follow-on patents demonstrate the value that will be introduced when a new infrastructure becomes available that transforms the world from rendered video to an instant, spaced, dynamic video assembly, allowing prompts to shape personalized content for every individual on the planet.
