WitnessAI launches Agentic Security to protect AI agents

Silicon Angle reports that WitnessAI Inc. has launched Agentic Security, a new capability designed to extend its AI confidence layer to autonomous and semi-autonomous AI agents, moving beyond large language models.

The new offering addresses the growing challenge of securing AI agents that can access internal systems, APIs, and third-party tools. Unlike traditional chat interfaces, these agents can act automatically, creating new attack surfaces vulnerable to prompt injection, data exfiltration, and unauthorized actions. Agentic Security aims to provide security teams with visibility and control by discovering agentic activity across various platforms, including ChatGPT plugins, VS Code AI extensions, and local agent frameworks like LangChain and AutoGPT. The platform analyzes tool usage to distinguish agentic sessions and maps accessed model context protocol servers. It also focuses on accountability by linking agent actions back to human users and capturing execution context, even when agents interact with each other.

Agentic Security introduces bidirectional runtime defense, scanning prompts to prevent attacks and tokenizing sensitive data. It also inspects agent outputs for policy compliance and harmful content. This development highlights the increasing need for specialized security solutions as AI workflows become more complex and integrated into enterprise operations.

Source: Silicon Angle