Generative AI already has use cases in security, including in video surveillance as well as in access control. Various challenges, such as large compute resource utilization and privacy concerns, still remain. This article looks at generative AI in access control and how certain challenges can be addressed.
Generative AI is a type of AI that can generate new contents in the forms of text, images, video or software code upon user requests. Increasingly, generative AI in security has evolved from a much-hyped concept to reality. “Development within the field of AI continues to race ahead. Deep learning technologies are the bread and butter of most analytics solutions within the security sector, while newer generative AI technologies are rapidly maturing. There is still a lot of hype in certain areas but real applications of generative AI in the security sector are becoming available,” said Axis Communications in its
post about security trends for 2025.
In the previous article, we discussed
generative AI applications in video surveillance. These include search efficiency enhancement and deep fakes detection. Access control, meanwhile, can benefit from generative AI as well.
“Generative AI can enhance access control by simulating potential bypass attempts, such as creating synthetic data of someone using a forged ID or mimicking authorized personnel. It can also strengthen decision-making in systems that combine multiple authentication factors. For instance, generative AI can synthesize behavioral patterns to refine systems that use gait analysis or voice recognition alongside facial scans, making it harder for bad actors to exploit gaps in access control,” said Will Knehr, Senior Manager of Information Assurance and Data Privacy at i-PRO Americas.
Knehr further said generative AI could improve biometrics by generating diverse datasets that include variations in facial expressions, lighting, and obstructions. “For example, it could create datasets of faces wearing masks or hats to train systems to identify individuals accurately in less-than-ideal conditions. Additionally, it can simulate spoofing attempts, such as 3D-printed masks or recorded voices, helping developers fortify biometric defenses against increasingly sophisticated attacks,” he said.
In fact, leading biometrics solutions provider ZKTeco has already embraced the latest technologies in AI, including GPT-4, a type of generative AI.
“GPT-4, with its impressive language understanding and generation capabilities, offers potential benefits for our solutions, such as improved user interaction, more intuitive system controls, and enhanced data analysis. We are optimistic that the integration of GPT-4 with our existing solutions will result in improved user experiences and more efficient system operations,” said Swift Wu, GM of the International Department at ZKTeco, in an
earlier interview with asmag.com.
However, Wu maintains that ChatGPT is just one facet of AI. “AI technology as a whole offers a broad range of opportunities to enhance our solutions, and we've been keen to leverage these opportunities. For instance, we use AI in our biometric algorithms, where machine learning techniques allow our systems to improve accuracy and performance over time. AI can also improve system integration, allowing different systems to work together more intelligently and efficiently,” he said.
Challenges remain
Despite its potential, generative AI still faces certain challenges in security. One is the fact generative AI algorithms require extensive compute power to run.
“Generative AI models are large and require much compute capacity to execute, which creates a debate in how to balance the cost of AI (both in terms of financial investment, but also in terms of energy use and environmental impact) with its value,” the Axis post said. “Generative AI can assist operators in interacting with security solutions in natural language but, for the foreseeable future at least, require significant processing power.”
“Key challenges include the computational resources required to train and deploy generative AI, compatibility with existing infrastructure, and concerns about adversarial use. For example, older video surveillance systems might struggle to integrate with generative AI algorithms due to limited processing power. Moreover, generative AI itself can be exploited by attackers to create convincing fake credentials or bypass biometric systems,” Knehr said. “Ensuring robust security measures and collaboration across teams is essential to address these issues.”
Privacy concerns are also significant, especially when generative AI processes sensitive data like facial recognition or video footage. “For instance, generating synthetic data might inadvertently reveal patterns or behaviors tied to real individuals, raising ethical questions. To address these concerns, organizations should implement strict data governance policies, anonymize sensitive information, and use encryption to secure data storage. Adhering to privacy laws like GDPR and maintaining transparency about AI applications are also vital steps,” Knehr said.