The AI Attack Lab: MCPwned and the Offensive Agent Cycle

John Z Black Apr 27, 2026 AI Security #mcp #ai-security #red-team #prompt-injection #ai-soc

Defenders are building AI SOCs. Attackers are already building the tools to bypass them.

This week saw the release of MCPwned, a Burp Suite extension designed specifically to audit (and exploit) the Model Context Protocol. MCP is the engine that lets AI agents talk to your local files and APIs. Most implementations were built for speed, not security. MCPwned lets you sit in that gap and inject prompts at the protocol level.

On the other side, the “Sable” methodology from Vulnetic is proving that offensive agents can now model “normal” traffic so effectively that behavioral detection systems don’t even blink.

We are shipping AI features faster than we can audit them. The offensive playbook is moving in machine time. If your security strategy relies on “AI-powered detection,” it is time for a rethink.

Check out the technical tutorial on MCPwned and learn why auditing the AI handshake is the next big defensive frontier.