Human-in-the-Loop Controls for High-Stakes Agent Actions
Blocking every action defeats autonomy; approving all defeats governance.
Madeleine Truchon
Contributing Editor, Secure Development
Madeleine Truchon
Madeleine Truchon holds an academic background in programming-language theory and spent several years auditing supply chains for open-source foundations before joining Threat Intel Review. She writes about how engineering choices made early in the software lifecycle shape an agent's attack surface at deployment.
10 stories
Incident Response Playbooks for Compromised AI Agents
Security teams need new playbooks because AI agent compromises leave no binary signatures to patch.
Madeleine Truchon
Secure Configuration Baselines for MCP Server Deployments
Protocol leaves three security gaps; a new benchmark fills them.
Madeleine Truchon
Credential Management Patterns for Tool-Calling AI Agents
Separating credential storage from agent execution prevents exploitation through compromised tokens.
Madeleine Truchon
Security Release Gates for AI Agent Deployments
Agents need security gates based on runtime behavior, not configuration checklists.
Madeleine Truchon
Proven Exploit Requirements for AI Agent Security Findings
Unproven findings mask real agent security risks that legacy frameworks cannot see.
Madeleine Truchon
Isolated Twin Environments for Agent Security Testing
Isolated twin environments catch multi-layer attacks that point-in-time testing structurally misses.
Madeleine Truchon
AI Agent Attack Surface Differences From Traditional API Threats
Agents expose attackers to trust boundaries that form at runtime, not design time.
Madeleine Truchon
Reconnaissance Techniques Targeting Deployed AI Agents
Attackers map AI agents' capabilities, memory, and boundaries before exploiting them.
Madeleine Truchon
Tool Poisoning Exploits Against MCP Servers
MCP servers can inject malicious instructions through tool metadata before any tool runs.
Madeleine Truchon