Training · AI security
Security training for LLM applications and AI agents
Developers attack an LLM application the way an attacker would, then fix what they found. The programme is based on the OWASP Top 10 for LLM Applications (2025) and the OWASP Top 10 for Agentic Applications (2026).
In short
- Who it's for
- Teams building LLM applications, agents or MCP servers, and the security champions in those teams.
- Length
- 1 or 2 days.
- Format
- On-site in Poland or Norway, or remote, in English or Polish, with a lab on an application I prepare or on yours.
- Price
- From PLN 7,000 net per training day. You order directly from me, with no agency in between, and the fixed price is agreed before we start.
Programme
- Day 1. AttackDirect and indirect prompt injection, meaning an instruction hidden in an email, a document or a web page the agent reads. Data exfiltration through tools, excessive permissions for agents and MCP servers. All of it in a lab on a working application.
- Day 2. Defence and processPermission design and tool isolation, a human in the loop for decisions that cannot be undone, security tests that run on every change. Plus coding with an AI assistant: secrets in code, made-up package names (slopsquatting), SBOMs and the supply chain.
We finish with a short look at regulation: what the Cyber Resilience Act, the AI Act and, for companies in Poland, the amended National Cybersecurity System Act (KSC, Poland's NIS2 implementation) mean for you. I run full regulatory preparation separately, as an advisory service.
How I know this
I designed and lead the development of Siemens Energy's internal industrial cybersecurity portal: OT and IT asset inventory, risk, vulnerabilities and update recommendations. At Prosit we built Chatmerce, a platform of agents that answer customers over RCS, SMS and the web in production, so I know the attacks in this training from our own systems. Prosit also has access in Anthropic's Cyber Verification Program (CVP), so we use Mythos 5.1 for defensive work on systems our clients own or maintain: triage, vulnerability verification and detection.
What you keep
A review checklist for LLM applications based on both OWASP lists, sample security tests for your CI, and the lab materials.
Questions
Do you also run security tests on our application?
Yes, as a separate service. I review the architecture and the permissions of tools and agents, then try to break the system through prompt injection, tool abuse and data exfiltration. You get a report with priorities. For a wider scope, such as infrastructure testing, Prosit's cybersecurity team joins in.
Can we practise on our own application?
Yes, after signing an NDA and on a test environment. The lab then doubles as a first security review. By default we practise on an application I prepare.
What level do participants need?
It is enough that they write code and know the basics of web application security. No pentesting experience needed.
Does the training cover the CRA, NIS2 and the AI Act?
Briefly, focusing on what the regulations mean for a development team. Full regulatory preparation for the company is a separate advisory service.
How much does the training cost?
From PLN 7,000 net per training day, plus VAT, ordered directly from me with no training agency in between. Typically 1 or 2 days. You get a fixed price before we start.
How to start
Book a short call and tell me who will be in the room and what your team struggles with most right now. I will tailor the programme to you, and you get a fixed price before we start.
Ask about training →Prefer to send a brief first? Email [email protected]. Or message me on LinkedIn.
Already have an LLM application or agent and want an outside look at its permissions and security before production? That is the production-readiness review.