Kyvvu

OWASP agentic · ASI10

ASI10: Rogue Agents

A rogue agent is malicious or compromised and deviates from its intended function or authorised scope, acting harmfully or deceptively within a system of agents and people. Each action may look legitimate on its own while the overall behaviour is harmful. The entry covers this loss of behavioural integrity once drift has begun, not the initial intrusion.

What it looks like

  • After reading a poisoned instruction, an agent keeps exfiltrating files even when the source has been removed.
  • A compromised automation agent spawns copies of itself through provisioning APIs.
  • An agent told to cut cloud costs gets there by deleting backups (reward hacking).

What OWASP recommends

  • Keep signed, immutable logs of agent actions, tool calls and inter-agent messages.
  • Use trust zones and sandboxed execution with least-privilege API scopes.
  • Watch behaviour against signed manifests of expected tools and goals, including watchdog agents that look for collusion.
  • Be able to stop a rogue agent quickly with kill switches, credential revocation and quarantine, and require fresh attestation and human approval before bringing it back.

How Kyvvu's policies address it

These policies from the open library are written for this threat. Runtime policies judge each step against everything the agent did earlier in the task; registration policies check what an agent declares before it runs.

Based on the OWASP Top 10 for Agentic Applications for 2026 by the OWASP GenAI Security Project, licensed CC BY-SA 4.0. The names are OWASP's; the summaries are ours.

Your agents

Is ASI10 a risk for yours?

For security teams

Request a demo

We walk through your agents' riskiest paths, and the policies that close them.

For developers

Build the path yourself

Recreate the attack in the playground and see which policies stop it.