Join our daily and weekly newsletters for the latest updates and exclusive content on industry-leading AI coverage. Learn More

Many people talk about AI as if it were magic, but Sohrob Kazerounian, a distinguished AI researcher at Vectra AI, is down-to-earth about it because he's building it every day. His team is developing a new generation of AI agents poised to take cybersecurity to levels beyond what humans can achieve alone.

In an era of increasingly sophisticated cyber threats, Vectra stands out by empowering organizations to detect and respond to advanced attacks in real-time. At the core of Vectra's innovation is attack signal intelligence, an AI-driven threat detection system that analyzes behaviors across hybrid cloud environments, networks and identities. Unlike traditional security solutions that rely on rules-based detection and laborious deep packet inspection, Vectra leverages AI to scrutinize encrypted and decrypted traffic metadata in a dynamic and holistic way.

Cybersecurity is a high-value domain for AI, but also a very demanding one. While success at individual tasks may seem impressive, Kazerounian points out that the difficulty compounds when agents are required to perform multiple actions in sequence. "Even if you have a 99% success rate at each step, as you take more and more steps, you have an exponential falloff to guaranteed failure," he said. In high-stakes domains like cybersecurity, healthcare and law enforcement, that isn't good enough.

To set the curve for reliable AI agents, Vectra is taking a multi-faceted strategy to development and operation, which includes:

A layered agent design;

Adhering to the novel security paradigms of attack signal intelligence (ASI);

Incorporating privileged access analytics (PAA);

Vectra takes a layered approach to agent design, using programmed constraints and guardrails and giving agents access to a knowledge graph that encodes a powerful store of curated expertise. Knowledge graphs (KGs) are massive data structures that represent facts about objects, people, places and the rich inter-relations between them. KGs are used by data-intensive projects such as Google Search. Vectra's KG ensures that each action taken by AI agents is backed by structured reasoning and validated against real-world ...