mpm

Michael McCord

Detection & response · Offensive research · Agentic AI for defense

A detection and response leader working where security meets applied AI. Today that means architecting AttackBench at NDay Security — an autonomous penetration-testing platform built on the assumption that both the model and its operator will sometimes be wrong.

Fifteen years got me here: de-anonymizing Tor clients at Carnegie Mellon's CERT, running global incident coordination inside Citi's Cyber Security Fusion Center, and leading the HUMAN Security R&D team whose detection system autonomously blocks over 200 million malicious bots a day while cutting analyst triage by 90%.

Equal parts architect and manager: I hire small senior teams, own the hard design calls, and stay deep in the stack rather than above it.

Open source

All repos →
TS offsec-labs-mcp An MCP server for managing your own authorized OffSec Proving Grounds (PG Play / Practice) labs from an MCP client: list and search the catalog, inspect machine details, start/stop/revert machines, and pull a machine's walkthrough into your session for personal study. 1 starAug 4, 2026 Py MetasploitMCP MCP Server for Metasploit pypi 3.1.01 starAug 1, 2026 Py nessus-export Export Nessus scan results from the command line via the REST API — with an automatic fallback that reconstructs a valid .nessus file even when the server's native export is locked (as it is on Nessus Essentials / trial installations). pypi 0.1.01 starJul 16, 2026 Py lamiai-oss Lamiai — interactive honeypot that draws an autonomous AI offensive agent into revealing intent, capability, and attribution. Based on Beezlebub. 1 starJul 8, 2026 SKILL.md antibot-recon-skill An agent skill for reconnaissance and reversing of antibotting platforms 2 starsJun 23, 2026

Writing

All posts →
AI security research · advisory Boundary-dependent guardrail visibility in tool-using agents A design weakness in where semantic monitors sit in agent pipelines, written up as an advisory: the necessary condition a control has to satisfy, a working proof of concept, an interactive coverage model, and a procedure for testing your own deployment. The remediation is not a better classifier. August 2026 Security research · open source Behavioral recovery from eager AI attackers AI attacks are scaling past what patching and detection can cover. But an AI agent's eagerness to finish its task is a weakness you can use: drop it into a decoy environment and read what it is after from what it does, no prompt injection to fail or tip it off. Real captures, cited threat reporting, and open code meant to go upstream into Beelzebub. July 2026 Security research Reversing anti-botting solutions has never been easier. Is PACT the solution? Cloudflare's new PACT proposal says inspecting browsers to catch bots has failed. I reverse-engineered a vendor that bets the other way, to test how much weight fingerprinting can still carry. June 2026