Cisco adds runtime protection for AI agents

Cisco has introduced Agent Runtime Protection in the Cisco AI Defense Python SDK, a new capability designed to secure AI agents as they move into production. The integration uses a single agentsec.protect() call to wrap LLM calls and MCP tool invocations with Cisco AI Defense inspection, applying checks to prompts, responses, and tool interactions without changing application code.

The protection targets risks that emerge when agents read untrusted content, call external tools, and act autonomously. Cisco cites threats including tool poisoning, indirect prompt injection, data exfiltration, and rug pull attacks, noting that traditional WAFs and API gateways are not built to understand LLM context or reasoning traces. According to Cisco’s AI Readiness Index 2025, 83% of companies plan to develop or deploy AI agents.

Agent Runtime Protection covers chat applications, agentic frameworks such as LangChain, LangGraph, CrewAI, AutoGen, Strands, Google ADK, and the OpenAI Agents SDK, as well as managed runtimes including AWS Bedrock AgentCore, Google Vertex AI Agent Engine, and Microsoft Azure AI Foundry. The SDK supports multiple providers, offers API and gateway modes, and can inspect MCP tools, prompts, and resources on both request and response paths.