Large-scale AI in video security survey: Addressing pain points across many verticals, but challenges remain

AI-enabled cameras remain one of the main drivers in the physical security and IoT market. The latest large-scale AI models, such as Guanlan by Hikvision, are at the forefront of this trend, as they revolutionize video security by developing a holistic understanding of video streams, and thereby enabling a host of new features such as video search based on natural language.
 
Adoption of AI has been increasing steadily over the past couple of years, and it is safe to say that many industry professionals have an eye on the next step—large-scale AI.
 
This is not surprising, as the benefits of the latest technology are impressive and plain to see.  Large-scale AI models, which use massive multimodal datasets and transformer-based architectures, can not only recognize objects and events, but also understand the relationships between them. This makes it possible for large-scale AI-enabled system to detect subtle anomalies or context shifts that would be invisible to rule-based or conventional AI approaches.
 
How long will it take, however, for the revolution in AI to take hold in physical security workflows in real-world applications?
 
To explore the adoption of large-scale AI in video security, asmag.com and Hikvision have teamed up to conduct a survey examining the deployment and potential of large-scale models. While the partner piece to this article explores user awareness, demand trends and adoption drivers, we will now dig deeper into pain points that users expect to see alleviated by large-scale AI, as well as verticals where they expect significant impact. We will also look into the challenges that remain to the widespread adoption of the latest AI models.
 

Key findings

Uptake of large-scale AI is already significant
As manufacturers are gradually adding the latest technology to their offering, such as Hikvision’s Guanlan large-scale AI powering DeepinView X cameras and AcuSeek NVRs, a significant share of professionals is already working with them or planning to adopt soon. More than half of the respondents in our survey, or 55%, say they are already using large-scale AI models, while another 20% say they are planning to start doing so within the next 12 months.
 
This finding, which suggests adoption rates might climb to about 75% within a year, matches the broader market trend toward rapid uptake. It also underlines the importance users give to staying up to date with regard to AI: In our survey, 52% say adopting large-scale AI is “very important for the future of video security,” while another 26% say it is “important.”
 
Adoption barriers, such as high cost (mentioned by 60% of respondents) and data privacy and compliance (mentioned by 57%) do not seem to slow down the trend significantly.
 

Large-scale AI addresses key pain points

Respondents strongly associate large-scale AI models with addressing long-standing pain points. When asked “What value do you see in the emerging deployments or applications of large-scale AI models in video security so far,” 73.4% cite improved operational efficiency, 62% say reduced human error, and 55.4% say better end-user experience.
 
The answers to the question “In which areas do you see the most need or potential for innovation powered by the latest large-scale AI?” similarly show that respondents expect advances in AI to automate or simplify the day-to-day operations of security teams—especially those that are time-consuming and prone to human error.
 
Respondents see the biggest potential for innovation in video search and forensic investigation (56%), object and event detection (39.1%), and real-time alerts and responsiveness (34.3%).
 
Video search and forensic investigation: Products powered by large-scale AI models, such as Hikvision’ AcuSeek NVRs, enable natural-language video search, making it possible to instantly retrieve relevant clips without manual review using pre-defined search terms.
Object and event detection: By distinguishing meaningful activity from background noise, as well as suspicious items from unsuspicious ones, large-scale AI powered cameras, like Hikvision DeepinViewX series, can cut false alarms by 90%,with much more accuracy compared to conventional AI cameras.
Real-time alerts and responsiveness: Large-scale AI powered context-aware alerts allow faster, more accurate reactions, streamlining decisions in critical moments.
 
Other top replies to the question about the highest potential for innovation were system setup and resource optimization (25.9%) and alarm configuration and filtering (22.8%). This shows that industry professionals also expect benefits beyond the daily workflows of security teams—in the initial setup of architectures and their refinement, for example.
 

High-impact verticals

Over half of the respondents, or 51.2%, see critical infrastructure as the vertical where “functionalities powered by large-scale AI models will become most common.” Over 10% ahead of the rest of the field of verticals, this shows that large-scale AI is seen as most beneficial in safety-critical, high-complexity environments where monitoring scale is large and error tolerance is low. A real-world example of this trend is the Jogja National Museum in Indonesia, where Hikvision’s AI-powered AcuSeek system enabled rapid incident search and intelligent alerts, cutting response times and enhancing artwork protection.  
Verticals in which respondents also expect significant adoption—such as transportation (40.3%) and industrial/manufacturing (38.5%)—are also characterized by high-scale deployments. Here, too, automated security operations powered by large-scale AI are especially beneficial.
 
Meanwhile, price-conscious verticals such as retail (21.6%) are seen as rather reluctant to take up large-scale AI. It is, however, noteworthy that answers were spread across all 10 verticals listed in the survey, with all of them ranging between over 50 and 20%. This shows that respondents expect large-scale AI to have an impact across the board—in some verticals sooner, in others later.
 

Adoption challenges: Regional differences over cost and privacy

The presumed high cost of large-scale AI-powered devices is seen as the main obstacle to the quick adoption of the technology. When asked “What challenges or obstacles do you see in adopting the latest large-scale AI models in security,” 60% cite costs, followed by data privacy and compliance concerns (57%), as well as integration with legacy systems (53%).
A closer look at the responses reveals regional patterns. Nearly four in every five respondents from Asia cite cost concerns (78.9%) as obstacles, while privacy and compliance issues tops the list among those based in Europe (EU and non-EU) even more clearly, with 81.1% citing them.
 

Skepticism receding?

Complexity and novelty of large-scale AI models are also a concern among respondents, albeit to a lesser degree. About a third of respondents (33.1%) cite “lack of technical understanding” as an obstacle, while 18% cite “resistance to change/ customer skepticism.”
 
As more users will come into contact with large-scale AI, this obstacle might drop down in the list, though. Organizations are widely embracing large-scale AI, despite the perceived challenges.
 
When asked “Which of the following statements best reflects your organization’s view on the latest AI,” less than a quarter say their employers have “no plans” to adopt large-scale AI (9.6%) or are “skeptical” about it (12.6%). In contrast, most respondents work in rather “AI-positive” organizations, which are only divided over when to take the step toward large-scale AI. While 38.5% say their employers are currently exploring the potential of the technology, 37.3% say they are actively testing or already using it.  

Conclusion

With offerings such as a series of new products powered by Hikvision’s Guanlan AI models, large-scale AI is no longer just a promise for the future—it is already reshaping the workflows and priorities of the video security industry. With Guanlan, Hikvision is applying large-scale AI across visual, language, and multimodal domains to address real-world needs such as perimeter protection, text search, and object detection.
 
Large-scale AI demonstrates strong generalization ability, allowing it to adapt across scenarios while reducing false alarms, enhancing analysis, and supporting more complex decision-making. Its strength lies in addressing persistent operational challenges while opening the door to new levels of automation and refinement.
 
The path ahead will be defined by how well the industry balances innovation with practical concerns such as cost and compliance. One thing is clear: large-scale AI is becoming a cornerstone of future security architectures and its full potential is only beginning to unfold, promising even broader possibilities in the years to come.
 
To explore the full survey findings and detailed industry insights on leveraging large-scale AI for video security, click here.