Valeri Milke
Founder & Principal AI Security Consultant @ VamiSec | AI Threat Modeling, LLM Security & Agentic AI
Valeri Milke is CEO of VamiSec GmbH and an information security and compliance expert focused on helping organizations translate European cyber and digital regulations into practical operating models. His work spans information security management, AI governance, secure development, and regulatory implementation across frameworks such as the EU AI Act, NIS2, DORA, CRA, and ISO/IEC 27001 and ISO/IEC 42001.
He advises organizations on how to operationalize regulatory requirements in real environments by connecting governance, engineering, risk management, and security architecture. His focus includes secure-by-design approaches, product and software security, AI risk management, third-party risk, and resilience requirements in regulated industries.
Valeri regularly works at the intersection of policy, technology, and implementation, with a particular interest in how European regulation affects software development, open technology ecosystems, and digital sovereignty.
Securing Agentic AI with PyTorch: Threat Modeling & LLM Red Teaming in Practice
Agentic AI systems are evolving from single-model assistants into distributed multi-agent ecosystems. These architectures introduce risks that traditional threat modeling often misses, including privilege escalation across agents, trust-boundary collapse, cascading failures, and resource starvation.
This session presents a practical methodology for threat modeling Multi-Agent Systems (MAS) using the MAESTRO framework (Multi-Agent Environment, Security, Threat, Risk, and Outcome). Rather than focusing only on prompt injection or model misuse, the talk shows how to analyze cross-layer risks created by agent autonomy, non-deterministic behavior, inter-agent communication, tool use, and chained authorization.
The session is aimed at application security engineers, security architects, AI engineers, and threat modelers who want a repeatable approach to securing next-generation autonomous systems.