Pillar Security has introduced a novel tool designed to address a critical blind spot in AI security: the complex attack surface created when AI agents are integrated with real-world business systems. RedGraph, a new product within its platform, moves beyond testing AI models in isolation to continuously map, discover, and validate exploits across an AI agent's entire operational environment, including its connections to databases, APIs, and internal data.
Pillar Security launched RedGraph, a first-of-its-kind AI attack surface mapping and testing tool.
It addresses the gap where traditional AI red teaming tests models in isolation, missing risks from integrated systems.
The tool interacts with live AI agents to map their entire operational graph and validate real exploits.
It uncovers both AI-specific risks (like prompt injection) and traditional app vulnerabilities exploited through AI.
RedGraph works with web-accessed AI agents built on platforms like Microsoft Copilot, Salesforce Agentforce, and Google Agentspace.
Findings feed into adaptive guardrails for immediate remediation, creating a continuous security loop.
Current AI security tools primarily focus on testing the underlying language model, which fails to account for the expanded risk profile when an agent is deployed. Once connected to business tools, databases, and permissions, AI agents create a new attack surface where business logic flaws and novel attack paths emerge. RedGraph is built to discover these risks by directly engaging with the AI agent in its runtime environment, simulating a real attacker's perspective.
The core innovation of RedGraph is its "graph-first" approach to attack surface management. It visually maps an organization's AI estate as a network of nodes (agents, tools, datasets) and edges (permissions, connections), revealing unintended relationships and potential pivot points for attacks. Crucially, it doesn't just identify theoretical vulnerabilities; it validates them by attempting real exploitation, proving the existence of a risk with demonstrable attack paths that engineering teams can directly address.
RedGraph is designed for the dynamic nature of AI systems. Its testing agents adapt in real-time, pivoting when blocked to find alternative attack vectors. This continuous assessment loop ensures that as AI systems evolve, their security posture is reassessed. Validated findings are fed directly into Pillar's adaptive runtime guardrails, enabling immediate hardening and transforming testing insights into active, evolving protection.
The launch of RedGraph signifies a maturation in AI security, shifting focus from the model to the operational system. By providing security teams with exploit-validated visibility into the complex attack surfaces of live AI agents, Pillar Security aims to close the critical gap between theoretical AI safety and the practical, integrated risks faced by enterprises deploying agentic automation at scale.
About Pillar Security
Pillar Security is a leading AI-security platform, providing companies with full visibility and control to build and run secure AI systems. Founded by experts in offensive and defensive cybersecurity, Pillar secures the entire AI lifecycle, from development to deployment - through AI Discovery, AI Security Posture Management (AI-SPM), AI Red Teaming, and Adaptive Runtime Guardrails. Pillar empowers organizations to prevent data leakage, neutralize AI-specific threats, and comply with evolving regulations.
TechIntelPro empowers B2B technology decision-makers with the latest industry trends, deep insights, leadership perspectives and actionable intelligence to drive both organizational and professional milestones—and thrive on the tech pulse.