It seems scarcely a day goes by when some new application of artificial intelligence doesn’t make headlines—whether it’s generating prose, paintings, or now, apparently, plausibly faked video footage. While the technology promises a raft of benefits for surveillance and incident analysis, it also raises the very real and increasingly urgent question: what happens to the integrity of security video in a world where any image can be convincingly fabricated?
In the age of generative AI, truth has acquired a glitch.
To help unpack the implications of this, I recently sat down with Leo Levitt, Director of System Integration Solutions at Axis Communications, and Chairman of the ONVIF Steering Committee—arguably one of the most influential voices in global video security standardization.
As Levitt pointed out, AI—generative or otherwise—isn’t inherently villainous. Like fire, it can warm or destroy depending on how it’s used. On the one hand, it’s enabled smarter video analytics, enhanced threat detection, and more efficient data processing. But on the other, it has flung open the gates to deep fakes and synthetic video creation tools so advanced, so accessible, that a high-schooler with a smartphone can generate a “security video” that could fool a seasoned investigator. And that, right there, is the crux of the issue.
Surveillance footage has long been the gold standard of incident verification—assumed authentic, relied upon by courts, insurers, employers and law enforcement alike. But with generative AI now capable of inserting people, removing others, and editing entire sequences in seamless 4K resolution, the question becomes not “Is this footage clear?” but “Is this footage real?”
In response, ONVIF is developing a promising countermeasure: video signing. This technology essentially embeds a digital watermark—think of it as a forensic birth certificate—directly into the footage at the point of capture. This signature includes the device ID, timestamp, manufacturer details, and a tamper-evident checksum, enabling downstream systems to verify whether the footage has been altered at any point in its life cycle.
It’s a subtle yet significant shift in thinking. Instead of chasing fakes, we verify truth. Levitt was quick to note that ONVIF’s goal isn’t to build a system that can always detect manipulation—an arms race no one can win—but rather, one that can irrefutably confirm authenticity.
If widely adopted, video signing could prove a much-needed antidote to the growing skepticism surrounding video evidence. But it also requires industry-wide alignment, regulatory support, and, perhaps most challengingly, a shift in end-user awareness. After all, what good is an integrity check if no one knows to look for it?
This is where the conversation takes a turn from the technical to the philosophical. Trust, once broken, is hard to re-establish. And if public faith in video evidence falters—if courts begin to doubt what they see on screen—the implications ripple far beyond security. Everything from traffic violations to workplace misconduct to acts of terrorism could become mired in doubt.
What’s needed now is not panic, but preparedness. Technology will continue to evolve—sometimes faster than we’re ready for. But if we want surveillance to remain a cornerstone of security, we must evolve too. That means embracing standards like video signing, demanding transparency from manufacturers, and educating clients and end-users about what “authentic” means in a digital age.
In the end, security has always been a dance between trust and proof. Generative AI just changed the rhythm. It’s time the industry learns a new step.
To hear the full interview, click here or subscribe FOR FREE to the Security Insider Podcast visit https://blubrry.com/asial_security_insider/
