Had a fantastic conversation today with Rob van der Veer as part of our AI Vulnerability Webinar Series discussing the role of the OWASP AI Exchange in securing agentic AI systems.
We’re entering a world where AI systems don’t just answer questions, they plan, decide, access tools, and execute actions. In many ways, we’ve given AI systems the keys to the building… before installing the locks.
A few key insights from the discussion: 🔹 The biggest risk isn’t just prompt injection, it’s how agents interact with tools and the real world. 🔹 Cross-user indirect prompt injection is an emerging threat many organizations are still underestimating. 🔹 The biggest security win today? Blast radius control through least privilege, oversight, and transparency. 🔹 Security teams must be willing to “spoil the party” with risk analysis before innovation turns into incident response.
What gives hope? Collaboration. Seeing organizations like SANS Institute and OWASP AI Exchange align on shared controls is exactly the kind of signal the industry needs.
Because securing agentic AI isn’t a tooling problem. It’s a guidance, architecture, and community problem and the work is just getting started.
#AIsecurity #AgenticAI #OWASP #OWASPAIExchange hashtag #AISecurity #CyberSecurity #LLMSecurity #AIgovernance #PromptInjection #AISafety #AppSec #SecurityCommunity