AI Agent Security
Prompt injection, MCP exfiltration, compromised agents. Red teaming, sandboxing, permissions, security hooks. What your CISO needs to know.
What you'll be able to do
Identify attack vectors specific to AI agents
Conduct agent red teaming: injection, exfiltration, escalation
Configure defenses: hooks, MCP sandboxing, permissions
Build an AI governance policy
Align with NIS2, OWASP LLM Top 10, ISO 42001
Program
Threats and attack vectors
- › OWASP Top 10 for LLM Applications (v2.0, 2025): critical risks
- › Direct and indirect prompt injection: live demos
- › MCP supply chain: malicious servers, data exfiltration, poisoned dependencies
- › Shadow AI: unapproved agents in the company
Agent red teaming
- › Workshop: attacking an agent via web content injection
- › Workshop: exfiltrating data via a trapped MCP server
- › Workshop: escalating an agent's permissions
- › Analysis: which defenses held, which failed
Defenses and secure architecture
- › Security hooks: validate every action
- › MCP sandboxing: allowlists, denylists, per-tool permissions
- › Granular per-agent and per-context permissions
- › Real-time monitoring: detect anomalies
Governance and compliance
- › NIS2 and AI agents: obligations
- › AI security policy: template
- › Agent and MCP server validation process
- › Response plan: what to do when an agent goes rogue
Practical info
2 days (14 hours)
CISO, CTO, DevSecOps, developers, security architects
Cybersecurity basics. AI agent understanding (AGT-01 or AGT-02 recommended)
4 to 10 people
30% theory, 70% offensive and defensive workshops on sandbox environments. Each participant leaves with an AI security policy.
A Pyrmetia co-founder — AI developer and cybersecurity expert.
Pricing
Post-training support
Related programs
MCP, Skills & Hooks — Agent Infrastructure
The core system. Connect Claude to your APIs (MCP), teach it methodologies (Skills), automate guardrails (Hooks). Build a custom agent ecosystem.
Agent SDK — Deploy Autonomous Agents in Production
The Claude Agent SDK is the same engine as Claude Code. Python or TypeScript. Custom MCP tools, programmatic hooks, multi-agent orchestration. We deploy a real agent in production.
AI Act & Autonomous Agents
The AI Act classifies autonomous AI agents as high-risk systems. Classification, obligations, documentation — and a workshop to build your compliance file. Deadline August 2026.
Request the full program
Bespoke program, tailored to your industry. First call is free.