Agents often run with a service account or a copy of the user's token that grants far more access than the current task needs. Attackers who influence the agent can use those privileges directly, and agents acting for one user can be tricked into reaching resources that belong to another.
Give each agent its own identity with short-lived, narrowly scoped credentials. Propagate the end user's authorization context to every tool call, never let agents reuse cached high-privilege tokens across sessions, and log which identity performed every action.
The same feature written the risky way, then with the fix applied.
How this risk has ranked in each OWASP release.
2026Latest
#03asi03-2026 Identity and Privilege Abuse
