2026-05-08 –, Building D/room 2, Workshops and AI Security Village
As AI agents evolve from simple chatbots into autonomous systems capable of executing code and making API calls, traditional security boundaries are failing. We can no longer rely on brittle regex filters or the "black box" safety rails of LLM providers. In this session, I will unveil the architecture behind the SovereignShield ecosystem a multi-layered, deterministic defense framework for modern AI applications. We will break down the engineering mechanics of our three core products: IntentShield (outbound action auditing), LogicShield (semantic enforcement), and the unified SovereignShield firewall.
This 35-minute technical session is an architectural deep-dive into the SovereignShield product suite, designed to show developers and security engineers how to mathematically secure AI endpoints.
We will cover the ecosystem in three distinct technical phases:
LogicShield: Securing the Cognitive Layer (10 mins)
Why traditional syntax filters fail against semantic attacks (like prompt injection and jailbreaks).
How LogicShield enforces deterministic logical boundaries on AI reasoning before an output is even generated.
IntentShield: Outbound Action Auditing (10 mins)
The danger of autonomous AI agents executing destructive API commands or exfiltrating data.
Deep dive into the ActionParser and Conscience modules. How IntentShield intercepts, audits, and blocks malicious intent at the execution layer.
SovereignShield: The Unified Firewall (10 mins)
Bringing it all together. How the core SovereignShield layer acts as a bidirectional proxy.
Live architecture breakdown of our 4-layer defense model (Inbound Input Filtering + Outbound Action Auditing) protecting a production API.
Conclusion & Q&A (5-10 mins)
How the community can integrate the SovereignShield suite into their own LLM pipelines today.
Mattijs Moens is an AI security researcher and the founder of SovereignShield, where he engineers deterministic, multi-layered defense architectures for autonomous AI agents. He is an active contributor to the OWASP AI Security and Privacy Guide (AISVS) and advocates strongly for open-source, independent oversight of AI training data. His current work focuses on building "IntentShield" and "LogicShield" semantic firewalls designed to intercept and neutralize AI-generated social engineering, prompt injection, and data exfiltration at the edge.