Discussion on "The OWASP LLM Top 10 2026 Is Not a Vulnerability List. It's a Blast Radius Document"

Mohit Kumar · 2026-08-24T09:09:14.200Z

Author: Mohit KumarPart of The Agentic Attack Surface series on securing AI systems, agentic workflows, and the AI software supply chain. Most people will read the 2026 edition of the OWASP Top 10 fo

A code graph tool measured net-negative on a small repo, so I removed it. That unused capability still made it an actor.

That's a really interesting point. I think that's one of the easy-to-miss parts of the actor vs. component distinction, capability itself can expand the blast radius even when the capability isn't being actively used. Removing the unused capability makes sense from a least-privilege perspective. I hadn't thought about measuring it from that angle, but I like the idea.

Excessive Agency climbing to number three matches what I'm seeing, the risk moved from what the model says to what it's allowed to do. Blast radius is the right lens, a wrong answer is cheap but a wrong action with tool access is not. How are you scoping agent permissions to keep that radius small?

Exactly! That's how I see it too. A wrong answer is often recoverable; a wrong action with access to real systems can have a very different impact. For permissions, I'm leaning toward treating the agent's capabilities as an explicit graph and looking at the possible tool combinations rather than evaluating each permission in isolation. The goal is to keep the agent's reachable blast radius as small as possible for the task it's actually supposed to perform. That's also something I'm exploring more deeply in my work on agent security.

Search Hashnode

The OWASP LLM Top 10 2026 Is Not a Vulnerability List. It's a Blast Radius Document

Responses4