What happened
On June 29, 2026, Cisco published Agent Runtime Protection in the Cisco AI Defense Python SDK — a one-line integration (agentsec.protect()) that uses dynamic code rewrites to wrap every LLM call and MCP tool invocation across chat apps, agentic frameworks (LangChain, CrewAI, AutoGen, OpenAI Agents SDK), and managed PaaS runtimes (AWS Bedrock AgentCore, Google Vertex AI Agent Engine, Azure AI Foundry) in AI Defense inspection for prompt injection and data exfiltration.
Why it matters
Covers all three layers of the agentic stack — direct LLM calls, orchestration frameworks, and cloud-managed runtimes — with no code changes beyond a single call. Directly addresses MCP tool poisoning, indirect prompt injection, and rug-pull attacks. Cisco's scale means broad enterprise reach for what is currently the most complete agent runtime protection available.
Applicability
Any enterprise deploying Python-based AI agents across LangChain, CrewAI, AutoGen, or cloud PaaS runtimes; evaluate immediately, especially where MCP servers are in use.