Deploying context-aware guardrails for AI agents

Traditional guardrails often struggle with enterprise AI agents due to context-blindness, frequently blocking legitimate user requests as false positives. Furthermore, standard content moderation filters fail to address the technical realities of modern agentic workflows, leaving systems vulnerable to multi-step social engineering and tool-chain exploitation. As organizations scale their generative AI deployments, the reliance on manual governance and generic taxonomies creates significant security and compliance gaps. In this live session, David Berenstein will present Giskard Guards, an independent European platform designed for context-aware, sovereign AI security. We will detail how technical teams can deploy policy-driven guardrails directly within their own infrastructure, ensuring that custom compliance frameworks are strictly enforced without compromising data sovereignty. Agenda: AI Guardrails 101: Understand the fundamentals of enterprise AI defenses, from mitigating data leakage and hallucinations to preventing complex prompt injections. Model vs. Agent Guardrails: Explore why standard content moderation filters fail in agentic workflows, and learn how to secure the full execution chain, including multi-step reasoning, tool calls, and parameter validation. Implementing Policy-as-Code: Transition from manual, static risk assessments to machine-enforceable OPA/Rego policies that can be versioned in Git and deployed directly on your infrastructure. Enforcing Regulatory Frameworks: Leverage pre-built, automated policy packs to ensure compliance with the EU AI Act and the OWASP Top 10 for LLMs, and learn how to add your own policies. Conclusion and Q&A. Who is it for: AI Product Managers, Heads of AI, AI/ML engineers, AI security professionals, Data scientists, and anyone building or deploying GenAI applications who wants to ensure their models are safe, reliable, and production-ready. Contact: https://www.giskard.ai/contact