Course · Training · Workshop
Agentic Red Teaming
Two days hands-on: simulate GenAI-assisted attacks on conventional systems, test your own agents and deliver evidence your defenders can use.
Generative artificial intelligence (GenAI) and agents based on large language models (LLMs) can support research, vulnerability analysis and attack-path planning. For red teams, this raises a practical question: which steps can you usefully delegate to agents, and where do experience, control and manual verification remain essential? In this course you build an agentic red team setup. You connect security tools through the Model Context Protocol (MCP), have agents develop hypotheses and check their results in the lab. The focus is on using agents to test conventional systems: web applications, identities and cloud services. A further focus is the security of LLM applications and agents themselves. Scope, rules of engagement, human approvals and a traceable audit trail are part of the setup from the start. You learn to distinguish plausible agent responses from demonstrated vulnerabilities and hand reproducible findings to your blue team.
What participants say
The course was highly educational. I discovered many tools and found the integration of these tools with AI into a developer environment extremely fascinating. I learned a lot about cutting-edge topics, which is especially impressive given the rapid evolution of the AI landscape. The knowledge I gained has already helped me improve and accelerate my own processes, significantly enhancing my daily workflow.
Cloud Platform Solution Architect, Helvetia Versicherungen Schweiz
* Voices from this or related courses.
These customers booked courses in the same topic cluster.More customers →
Content
This training is for engineers who want to reproduce GenAI-assisted attacks with traceable evidence and use them to test their defenses. Exercises combine conventional targets with the additional trust boundaries of agent applications.
The course consists of the following topics and can be adapted to the audience. For in-house courses we agree on focus areas such as web applications, Active Directory, cloud or containers. Within two days, we examine selected attack paths in depth.
Objectives and test boundaries
- Define objectives, scope and rules of engagement
- Plan approvals, data handling, stopping conditions and recovery
- Hands-On: create a lab test plan with measurable success criteria
The agentic red team stack
- Agent planning, tool use, context and memory
- Connect security tools via MCP; restrict permissions and target systems
- Cloud and local models: handling findings and confidential data
- Hands-On: set up an agent harness with logging and human approvals
Recon and attack surfaces in the lab
- Map assets, services, identities and trust relationships
- Interpret scan results and develop attack-path hypotheses
- Identify hallucinations and false positives through independent checks
- Hands-On: investigate a lab application and justify the next tests
Testing conventional systems with agent support
- Verify vulnerabilities and misconfigurations in web applications and cloud services
- Examine identity and permission boundaries along an attack path
- Compare manual tests with agentic workflows: quality, time and limitations
- Hands-On: reproduce and document an authorized attack path with synthetic data
Assessing GenAI-assisted social engineering
- Phishing and synthetic voices as scenarios for authorized exercises
- Evaluate technical controls alongside approval processes
- Workshop: plan a simulation with fictional identities and clear success criteria
LLM applications and agents as test targets
- Direct and indirect prompt injection through documents and tool responses
- Data leakage, excessive permissions and missing boundaries between users
- Distinguish model responses from actual access to data and tools
- Hands-On: test a lab agent with synthetic data, harden it and retest
Findings and collaboration with the blue team
- Document reproduction steps, evidence, impact and remediation
- Compare attack steps with telemetry and detection gaps
- Hands-On: run the final scenario, hand over findings and retest countermeasures
At the end of the two days you have an agentic red team setup, a traceable test plan and a findings report from the lab. You know which tasks you can delegate to agents and which results you need to verify yourself.
The actual course content may differ from the above depending on the trainer, delivery, duration and the composition of participants.
Request this course in-house
Request a public date
No suitable public date? Register without obligation — once there is enough interest we schedule a new public date and let you know first.
More about Agentic Red Teaming
Agentic red teaming combines offensive security testing with agents that operate tools and propose investigation steps. Its value comes from verifiable results: an agent suggestion is not yet a finding. Reproducible evidence, actual impact and a retest after remediation are what count.
The course covers both sides: agents as tools for testing conventional systems and agents as test targets themselves. For the latter, we examine whether untrusted content can influence agent behavior and which boundaries protect data and tool access.
Related courses
- Agentic Blue Teaming: detection, triage and incident response with agents.
- Defense in the Age of Agentic Attacks: priorities, organization and crisis readiness for leadership.