AI agent security training
The risks of autonomous agents (prompt injection, exfiltration via MCP) and their countermeasures: sandboxing, permissions, security hooks. AI agent red teaming.
Target audienceCISOs, CIOs, DevSecOps, developers, security architects
What you'll be able to do
Identify attack vectors specific to AI agents
Implement sandboxing and permission management mechanisms
Configure security hooks to detect and block malicious behaviour
Conduct a red teaming exercise on an AI agent
Define a security policy for deploying agents in production
Programme
- Day 1, morning
The AI agent threat model
- The 3 attack surfaces: prompt, tools, context
- Direct and indirect prompt injection: live demonstrations
- Injection vectors: web pages, emails, documents, API responses
- Defence techniques: data/instruction separation, validation, human confirmation
- Day 1, afternoon
Exfiltration, sandboxing and permissions
- Exfiltration via agent tools: files, URLs, APIs
- MCP and attack surface: malicious servers, tool poisoning, escalation
- Sandboxing: filesystem, network, process isolation
- Workshop: configuring a secure environment for an agent
- Day 2, morning (optional)
Security hooks and red teaming
- Detection hooks: secrets, suspicious URLs, dangerous commands
- Prevention hooks: block push --force, limit outgoing requests
- Red teaming in pairs: attacking and defending an agent
- Hook bypass: testing protection robustness
- Day 2, afternoon (optional)
Security policy and monitoring
- Writing your organisation's agent security policy
- Production monitoring: what to log, KPIs, alerting
- Incident response: detection, containment, investigation, recovery
- Action plan: the first 5 hardening actions
Practical information
1 day (6 hours) or 2 days (12 h 30), excluding breaks
CISOs, CIOs, DevSecOps, developers, security architects
Basic IT security knowledge. Familiarity with AI concepts.
1 to 10 people
30% theory, 70% offensive and defensive workshops on sandbox environments. Each participant leaves with an AI security policy and configured hooks.
Entry assessment, assessed red-teaming exercises. At the end, a before-and-after self-assessment of skills, then the trainer validates each learning objective (acquired, in progress, not acquired), recorded in the end-of-training certificate.
Ulysse Trin, founder of Colombani.ai, AI and security consultant, holder of the Cyber Solutions and Cyber Deployment Practitioner certifications issued by OpenAI.
At least 2 weeks between enrolment and the start of training.
Qualiopi-certified provider (certificate QUA009472), registered under training provider number 11757549975 (this registration does not constitute State approval): OPCO or FIF-PL support can be requested. The decision rests with the funding body, based on your sector, its criteria and its budgets. Colombani.ai prepares the application with you.
From €1,800 excl. VAT per day for an in-company group; final price depends on group size and length, on quote. OPCO or FIF-PL funding can be requested depending on your status; the amount left to pay depends on the sum granted by the funding body.
Receive the programme and a quote
Tailored programme and quote within two business days.
Accessibility
This course is accessible to people with disabilities. Contact the disability officer in advance to discuss possible adjustments.
Ulysse Trin, referent-handicap@colombani.ai
Post-training support
Frequently asked questions
Do I need pentesting skills for this training? +
No, but basic IT security knowledge is needed. The training is progressive: we start with the threat model before moving to red teaming.
Is red teaming done on real agents? +
Yes, on agents configured in a sandbox for the exercise. You attack and defend real agents in a controlled environment.
Does this training cover MCP risks? +
Yes. MCP risks (malicious servers, tool poisoning, privilege escalation) are covered in the session on data exfiltration, and MCP server permissions in the session on sandboxing. It's a critical attack vector for agents.
Related courses
Advanced Claude Code training: hooks, subagents, MCP and continuous integration
Two days on your repositories to take Claude Code to production: security and quality hooks (PreToolUse, PostToolUse), orchestration of subagents and parallel sessions on worktrees, MCP servers connected to your tools, headless Claude Code in continuous integration (GitHub Actions), permissions and defence against prompt injection.
AI compliance training: AI Act & GDPR
Classify AI uses, document transparency and risk controls, and build a roadmap aligned with the revised 2026-2028 AI Act timetable.
Receive the programme and a quote
Tailored programme and quote within two business days.
Got a project in mind?
Describe your situation. The first 30-minute call is free. Get a frank response within one business day.