Nadav Cornberg, CEO of Eve Security, highlights the need for real-time security for agentic AI in healthcare environments to enforce policies and map agentic activities within an organization. There are unique risks posed by AI, such as ambiguity from natural language commands, the potential for unintended or malicious actions, and the amplified impact of a single AI error compared to human error. Nadav points out that many hospitals are likely unaware of the extent to which agentic AI is already active in their systems, and of why conducting risk assessments and policy reviews should be a priority.

Nadav explains, "Eve Security provides two main services. One is our runtime security solution, which allows us to enforce policies when you connect any type of AI agent to a critical data source or system. The other solution that we provide is our AIDR, where we give you a full topology of what agentic activities are running in your organization. We do that by connecting to security systems like an EDR, NextGen Firewall, or your SIM."

"Just like we'll see a manager in the employee environment, in the physical world, that's why we have managers as well to guide employees on the work they want to do. How that works is we sit either on top of existing gateways or proxies, and that's how you will connect those agents to critical systems, or we connect to hooks."

"At the end of the day, the reality of how we're communicating and engaging with agents brings that new necessity. Existing security tools do not deal well with natural language. In addition to not dealing well with natural language, they're not doing well with trying to understand the true intent behind an action, and it's more built on static parameters. And that's the real gap that introducing AI agents into environments has brought. The risks are ambiguity and the unpredictable, non-deterministic behavior."

#EveSecurity #AgenticAISecurity #SecurityatRuntime #AISecurity #HealthcareAI #PatientSafety #HospitalSecurity #AgenticAI #CyberSecurity #HealthIT #DataProtection

Eve.security

Download the transcript here