OWASP agentic · ASI03
ASI03: Identity and Privilege Abuse
Agents often act with credentials, tokens or sessions that were delegated to them, and they pass work to other agents. Attackers exploit these delegation chains, inherited roles, cached credentials and default trust between agents to gain access they should not have. Without a distinct, governed identity per agent, enforcing least privilege and attributing actions becomes very hard.
What it looks like
- A manager agent hands its full permissions to a narrow sub-agent, and a manipulated query uses them to pull HR and legal data.
- An agent caches admin credentials from an earlier task, and a later, less privileged user prompts it to reuse them.
- A low-privilege or fake agent (for example a forged 'Admin Helper' agent card) relays instructions that a high-privilege agent executes without re-checking the original user's intent (confused deputy).
What OWASP recommends
- Issue short-lived, task-scoped credentials per agent and per task, bound to subject, resource, purpose and duration.
- Isolate agent identities, sessions and memory, wiping state between tasks.
- Re-authorize each privileged step through a central policy engine and require human approval for privilege escalation or irreversible actions.
- Monitor for permissions gained through delegation chains and for token use outside the intent it was issued for; manage agents as non-human identities in IAM.
How Kyvvu's policies address it
No policy in the open library is mapped to ASI03 yet. See the policy library for what it covers today.
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 ASI03 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.