CyberWiseCon Europe 2027 in Riga: Valeri Milke speaks on threat modeling and agentic AI red teaming
October 31, 2026
Talk accepted: VamiSec at CyberWiseCon Europe 2027
Valeri Milke will speak at CyberWiseCon Europe 2027, taking place in Riga from May 26 to 28, about threat modeling and agentic AI red teaming. The conference's Content Selection Board has selected our talk “Securing Agentic AI: Threat Modeling & Agentic AI Red Teaming in Practice” for the onsite programme. Valeri Milke is founder and managing director of VamiSec GmbH. The conference takes place at Apollo Kino Akropole, Latgales iela 257; it is preceded by a separate workshop day at the National Library of Latvia on May 25. The talk runs for 45 minutes including Q&A and will be given in English. We will add the day and time as soon as the organiser publishes the programme.
CyberWiseCon Europe: three conferences, one registration
CyberWiseCon Europe is an IT security conference run by the organiser DATA MINER and moves to Riga in 2027. It takes place together with the developer conferences DevDays and DevOps Pro Europe; one registration covers all three. The organiser expects more than 700 attendees and over 100 speakers from more than 35 countries across six tracks and more than eight workshops. Topics include AI and machine learning threats, ethical hacking, ransomware, zero trust, cloud defences and social engineering. Programme and tickets: https://cyberwisecon.eu/
What the talk is about
AI agents read enterprise data, invoke tools, retain memory and delegate tasks. This expands the attack surface beyond the model: untrusted content can influence an agent's plan and trigger an unauthorised action when trust and authorisation boundaries fail. The talk connects architecture-based threat modeling with pentesting and red teaming for agentic AI. MAESTRO structures the assessment of agents, prompts, tools, memory, orchestration and identity boundaries, while STRIDE supports the analysis of conventional application components and data flows. The OWASP Top 10 for Agentic Applications and for LLM Applications provide the risk vocabulary, the OWASP AI Testing Guide informs test design, and selected OWASP AISVS requirements define the verification objectives.
From attack path to evidence: the worked example
Using an illustrative retrieval-and-tool workflow with synthetic data, the talk follows an indirect prompt injection towards a proposed privileged tool call. It examines tool and parameter authorisation, binding approvals to the specific action, least privilege and direct backend access. Positive controls and negative tests show whether a safeguard works or whether the test environment is simply broken. Correlated policy decisions and backend effects show whether a protected action was actually prevented.
What participants take away
The result is a repeatable workflow: identify a trust boundary, formulate an attack hypothesis, design a bounded test, verify the control and retain useful evidence. Participants learn to structure an assessment of agentic applications around MAESTRO with supporting STRIDE checks, to translate selected threats into bounded red teaming and pentesting tests using OWASP risk categories, and to connect policy decisions and backend effects to verification objectives, remediation and retesting. The talk is aimed at security engineers, penetration testers, red teamers, application security practitioners and developer teams who build or assess agentic applications. It is vendor-neutral and contains no product pitch.
Our practice behind it
What we show in Riga is everyday project work for us: we model agent systems with MAESTRO, combine STRIDE with the OWASP catalogues and use agentic AI pentesting and red teaming to test which modelled risks are actually exploitable. A recording of an earlier talk by Valeri Milke on the topic is available on the PyTorch YouTube channel: https://www.youtube.com/watch?v=vvoDruv681A · AI threat modeling with MAESTRO: https://vamisec.com/en/ai-threat-modeling-maestro · Pentesting AI agents: https://vamisec.com/en/agentic-ai-pentesting · AI red teaming knowledge page: https://vamisec.com/en/wissen/agentic-ai-security/ai-red-teaming · Threat-model-guided security testing: https://vamisec.com/en/wissen/agentic-ai-security/threat-model-guided-security-testing
Meet us in Riga
We look forward to exchanging ideas with the security and developer community onsite. If you would like to talk about agentic AI security, threat modeling or AI red teaming in Riga, contact us at contact@vamisec.com. All VamiSec talks are collected in our news: https://vamisec.com/en/news?category=talks
All links
CyberWiseCon Europe 2027: https://cyberwisecon.eu/ · Organiser FAQ: https://cyberwisecon.eu/faq/ · MAESTRO by the Cloud Security Alliance: https://labs.cloudsecurityalliance.org/maestro/ · OWASP Top 10 for Agentic Applications: https://genai.owasp.org/resource/owasp-top-10-for-agentic-applications-for-2026/ · OWASP Top 10 for LLM Applications: https://genai.owasp.org/llm-top-10/ · OWASP AI Testing Guide: https://owasp.org/projects/ai-testing-guide · OWASP AISVS: https://owasp.org/projects/artificial-intelligence-security-verification-standard-aisvs-docs · Valeri Milke on LinkedIn: https://www.linkedin.com/in/valeri-milke/
Do you have questions about your organization's IT security?
Free Initial Consultation →