From Surveillance to Intelligence

• By Fredrik Nilsson
• Nov 24, 2025

For years, AI use in the physical security field was limited by factors like computing power, image quality, bandwidth and cost, leaving consumers unsure when (or whether) to buy into “the hype.” But with breakthroughs in areas like edge technology, video compression, and cloud architecture, that is beginning to change.

As AI capabilities grow more advanced, we are seeing greater intelligence embedded directly at the network edge, enabling real-time decision-making, reduced bandwidth and storage consumption and faster incident response. Seamless integration between edge devices, on-premises systems, and cloud platforms is also unlocking new efficiencies and allowing security teams to leverage actionable insights to approach their responsibilities in a more proactive manner.

But as AI adoption continues to accelerate, security providers are faced with new security and privacy challenges—not to mention regulatory and ethical considerations. Businesses are beginning to think more carefully about not just getting the most out of their AI solutions but doing so in a manner that prioritizes long-term accountability and sustainability. To understand the future of AI in physical security, it is important to understand how it has evolved over the years—and how advancements at the network edge have made AI more accessible than ever to today’s organizations.

Early Analytics and the Emergence of AI
One of the most familiar challenges when it comes to AI is computing power. Consider the conversations surrounding today’s generative AI solutions. Responding to text queries requires a significant amount of power. When AI tools are asked to analyze images, the power requirements expand even further. And when it comes to video, the increase is exponential. Since security teams deal mostly with video technology, that poses a significant problem for organizations that want to take advantage of AI-powered analytics.

Until recently, organizations had two bad (or at least expensive) options. On the one hand, they could invest in servers and GPU clusters to expand their on-premises computing power — but that would be prohibitively expensive. On the other hand, they could upload their video to the cloud for analysis — but that would also incur unreasonable storage and bandwidth fees.

As a result, only well-resourced organizations could afford to leverage AI in their security operations, and — even then — in a limited form. It was not until the arrival of today’s advanced edge devices, equipped with more powerful chipsets and enhanced video compression capabilities, that AI-based solutions became affordable and accessible to a broader range of businesses.

Powerful Edge Devices Are Changing the Game
That change began with advancements on the edge. Thanks to modern chipsets, many video devices now come equipped with deep learning capabilities — and some level of built-in AI has become standard. Deep learning allows these devices to filter out important data, identifying objects and determining which are important or relevant.

That does not just mean detecting obvious security concerns — these devices can collect vast amounts of data for later analysis, logging metadata and images that a Large Video Model (LVM) can automatically classify, significantly enhancing search functionality. That even applies to unusual use cases that go far beyond what traditional surveillance systems are capable of — and, most important, it does this before the video ever leaves the device.

That makes a significant difference. Uploading metadata (and short, accompanying video clips or images) to the cloud requires far, far less bandwidth and storage than uploading full-length videos, especially with recent advancements in video compression. Because today’s cameras can pre-filter objects like people, vehicles, or animals and identify characteristics like plate number, car color or shirt color, organizations can take advantage of that data to enhance their security and safety capabilities.

For example, if a child becomes lost in a retail store, the security team no longer needs to scrub through video manually. Instead, they can use AI at the edge to filter relevant video and feed those images to the LVM, which will return any video frames with objects that match the child’s description based on height, brand of sneakers, or other characteristics.

Years ago, it would have been significantly more expensive to run an analytic like that — requiring a custom-built solution with burdensome infrastructure demands — but modern edge devices have made it accessible to everyone. It also saves time, which is a critical factor if a missing child is involved.

Video compression technology has played a critical role as well. Over the years, significant advancements have been made in video coding standards — including H.263, MPEG formats, and H.264—alongside compression optimization technologies developed by IP video manufacturers to improve efficiency without sacrificing quality. The open-source AV1 codec developed by the Alliance for Open Media—a consortium including Google, Netflix, Microsoft, Amazon and others — is already the preferred decoder for cloud-based applications, and is quickly becoming the standard for video compression of all types.

AV1 both enables a substantial bitrate reduction and reduced video storage needs, which means organizations can use more devices and analyze more video without breaking the bank. Since AV1 is already used by most major cloud providers, it is easy to deploy and makes integration easier than traditional video compression codecs, making it perfect for modern use cases. And since technology is open-source and royalty-free, it further enhances the accessibility of modern video technology — including the AI-based video analytics organizations are increasingly relying upon.

Ethical Considerations Are Not Always Obvious
While the growing accessibility of AI-powered video analytics offers substantial benefits, it also invites ethical concerns. Technologies like facial recognition can be incredibly useful for enhancing security and operational efficiency, but—like virtually any tool—they carry the potential for misuse if not implemented thoughtfully. That is why it is important for both vendors and their customers to prioritize security, privacy and other considerations.

Regulatory and compliance bodies have stepped in here to provide helpful guidelines, and many organizations have shifted their attention to analytics like facial detection or face matching, which provide similar functionality while limiting the potential for misuse. While no technology is inherently “good” or “bad,” vendors do have a responsibility to reduce the likelihood of negative outcomes. That includes protecting personal data, which is why many manufacturers are now placing greater emphasis on cybersecurity measures. No one wants to suffer cyber breach because of a physical security device.

Other ethical concerns are less obvious. Overpromising, unrealistic expectations and incorrect assumptions have become a concern when it comes to AI-based solutions, especially since many customers may not have a high degree of technical expertise. This knowledge gap has created opportunities for some providers to obfuscate their actual capabilities.

For example, some may claim to upload all video to their LVM for analysis, which sounds great — but results in the same excessive bandwidth and storage usage that slowed the adoption of early analytics. Additionally, if the vendor relies on a public AI service, customer surveillance video will certainly be used as training data for the model, which can create significant security risks. Other vendors may claim that their cameras can carry out all necessary analysis without sending video to the cloud. While it may sound great to some consumers, this, too, is spurious at best — a camera with such a massive GPU load would be outrageously expensive, difficult to scale, and prone to overheating.

Transparency is critical here, especially on the part of device manufacturers. It is important to be honest about the actual capabilities of a device — because, while overpromising may deliver a quick sale, it will result in disappointed (and disillusioned) customers. Technology demos and pilot programs can be valuable tools to set realistic expectations by allowing customers to experience firsthand how technology performs in their specific environments. Today’s AI-based analytics can deliver tremendous value to businesses from both a security and operational perspective, but exaggerating their capabilities can lead customers to doubt the efficacy of the technology.

By being as transparent as possible, vendors can help provide customers with the context they need to understand the current state of the video market — and the reason edge devices play such a significant role.

As a result, being transparent is not just the right thing to do — it can actively build trust with potential customers and help them avoid the exaggerations and overpromises that plague the security market.

Recognizing the Role of the Network Edge
Scalability has always been a challenge in the security industry — how can organizations deploy, manage, and monitor the devices they need to secure their locations effectively without allowing their budgetary requirements to become unsustainable?

Thanks to modern edge devices, AI-based analytics are now helping them answer that question more easily. Today’s analytics are helping operators conduct pre-analysis, identifying objects and images and feeding the relevant metadata into an LVM before the video even leaves the devices. Security teams can now rely on accurate and dependable automated alerts to notify them of security events in real time, all while leveraging the data from their devices to produce actionable insights that can improve future operations.

With devices growing more powerful, compression technology becoming increasingly advanced, and AI capabilities more widely adopted, the security industry stands on the cusp of a transformative era. As we venture further into the age of AI, it is important to recognize the technology supporting the industry on its journey—and the critical role that the network edge continues to play.