Agentic AI Security: Evolving with Speed, Access & Scale

Security for agentic AI needs to evolve as fast as the AI itself. That's the view from Eyal Alkalay, CIO at Verbit. He shared his thoughts with CTech on the real risks of agentic AI. Here's the thing: speed, access, and scale are the main concerns, not slower adoption. An AI agent can take many actions across systems in seconds. If it has too much access, receives bad instructions, or makes a mistake, the impact can grow very quickly. What's interesting is that Verbit doesn't believe slowing down AI adoption is the answer. Instead, the challenge is to ensure security, identity, permissions, and monitoring keep pace with the agents. Verbit actively uses and tests agentic capabilities, but human oversight is still required for high-risk activities. They approach AI agents like other privileged access, scoping access, limiting permissions, and logging activity. The bottom line: AI adoption is moving fast, creating new risks, and companies are working to evolve their security models to keep up. This matters because it highlights the ongoing challenge of safely integrating advanced AI into operations.