PRE Security Launches AgentGuard to Secure AI Agents

New Agentic Detection, Prediction, and Response (ADPR) capability spots AI agents drifting off mission and blocks damaging actions before they execute

PRE Security, an AI Native cybersecurity company, today announced PRE AgentGuard, an ADPR capability within the PRE Security Platform that helps enterprises monitor and control the autonomous AI agents operating across their environments.

AI agents now write code, execute commands, use credentials, query databases, and call APIs, often at machine speed with little oversight. That raises a new security dilemma: what happens when an agent is tricked, compromised, makes a mistake, or starts doing something it was never meant to do?

AgentGuard treats each agent as a persistent identity with a mission, a behavioral history, and an expected pattern of activity. Rather than inspecting isolated prompts or actions, it continuously evaluates whether an agent is still on mission, what it will likely do next, and whether that action should be allowed.

Also Read: Menlo Security Closes Zero-Day and AI Agent Attack Gaps with Google Security Operations Integration

Behavioral Analytics Built for AI Agents

At the core of AgentGuard is AIBA (AI Agent Behavioral Analytics), which extends the UEBA model security teams already rely on to AI agents, evaluating each agent’s identity, mission, tools, resources, and sequence of actions to detect deviation and dangerous trajectories.

AgentGuard sits inline between an agent and the tools it uses. It scores each action against the agent’s mission and history and enriches the decision with PRE Threat Intelligence, so an agent contacting a known-malicious domain is treated very differently from one calling a trusted service. At the point of execution, AgentGuard can allow, monitor, require human approval for, or block the action.

Also Read: IgniteUps.ai Launches NEXUS for Automotive AI Management

Seeing the Trajectory, Not Just the Action

Consider an accounts-payable agent whose normal workflow is to read an invoice, validate the vendor, check the purchase order, and prepare payment. If manipulated, its activity may shift to accessing a credential store, enumerating sensitive data, and attempting an external transfer. Each step may look legitimate; together, they describe exfiltration in progress.

PRE has demonstrated this scenario: AgentGuard recognized the agent’s trajectory and blocked the external transfer before any data left the environment. In another demonstration with an AI coding assistant, the same system command was permitted during normal operation, but restricted once the agent’s preceding behavior became suspicious.

Also Read: AYTM Launches Bridge for AI-Powered Research

“Prompt filters ask whether a prompt is dangerous. AgentGuard asks whether an agent is becoming dangerous,” said John Peterson, Co-Founder and CEO of PRE Security. “Instead of telling an analyst that an agent exfiltrated data, we can tell them it was heading toward exfiltration, and that we stopped it before the data left.”

Extending Predictive Security to the Digital Workforce

AgentGuard extends PRE Security’s mission to predict and prevent cyberattacks into autonomous AI. PRE is also integrating AgentGuard with its Log2NLP™, FAST autonomous investigation, and Agentic Surveillance technologies so suspicious agent behavior can be stopped, investigated, and continuously monitored on one platform.

Write to us [⁠wasim.a@demandmediaagency.com] to learn more about our exclusive editorial packages and programmes.