Skip to content
AI

Why can a junior agent in my pipeline silently inherit my top-level authority?

84

Opportunity

When you build a multi-agent pipeline where an orchestrator delegates to sub-agents, there is no structural enforcement of privilege levels between them. A retrieval agent that processes untrusted web content sits one hop from an orchestrator that holds admin-level tool access, with nothing preventing the retrieval agent's output from carrying instructions the orchestrator will treat as authoritative. Research on LangGraph pipelines found that shared-memory poisoning propagates to execution in every undefended trial, and that a compromised sub-agent can cause the orchestrator to execute actions far beyond the sub-agent's own permissions. No production multi-agent framework ships privilege-separated hierarchies as a default, and developers who want containment must implement it themselves from scratch.

Why it matters

Vertical privilege escalation through agent hierarchies will become the dominant attack surface as multi-agent pipelines replace single-agent workflows in production.

How I score the opportunity

The Opportunity Score is my own read, not a measurement: how much it hurts, how often it bites, and how little exists to solve it today. Higher means I think it is more worth building.

Severity9/10

How much pain it causes when it shows up.

Frequency7/10

How often people actually run into it.

Whitespace8/10

How little good tooling exists for it today.

More problems worth solving