How to resolve
Add callbacks before/after agent calls or input guardrails on delegating agents. Validate delegated tasks against a scope allowlist. Limit delegation depth.Risk
Governance/Compliance This vulnerability falls under ASI10:2026 — Rogue Agents in the OWASP Top 10 for Agentic Applications. Compromised or misaligned agents act outside their intended scope without external triggering. Security Agent can delegate tasks to sub-agents with no scope, identity, or depth constraints. If exploited, this can compromise the agent’s integrity, confidentiality, or availability, potentially affecting downstream systems and data.Explanation
Agent can delegate tasks to sub-agents with no scope, identity, or depth constraints. This falls under ASI10 (Rogue Agents): Compromised or misaligned agents act outside their intended scope without external triggering.Specifications
Trigger- Agent scan detects agent can delegate tasks to sub-agents with no scope, identity, or depth constraints
- allow_delegation=True with no whitelist
- GroupChat with no speaker selection policy
- recursive agent spawning with no depth limit
- HIGH