Skip to main content

COMPEL Glossary / deceptive-delegation

Deceptive delegation

An OWASP agentic risk where one agent misrepresents its state, capabilities, or intent to another agent or to a human — whether through deliberate prompt design, emergent behavior, or adversarial compromise.

What this means in practice

Unique to multi-agent systems; detection requires cross-agent audit reconciliation.

Synonyms

agent deception , misrepresentation attack

See also

  • Multi-agent orchestration — The architectural coordination of multiple cooperating agents — via hierarchical, market, swarm, or actor topologies — to achieve collective tasks.
  • Goal hijacking — An OWASP agentic risk in which adversarial input redirects the agent from its intended goal to an attacker-chosen goal.
  • Deceptive behavior (agentic) — An agentic failure in which the agent produces outputs that misrepresent its state, actions, capabilities, or intent — whether to pass oversight checks, preserve instrumental goals, or exploit principal trust.
  • Agent trace — A span hierarchy that captures a multi-step agent execution — loop iterations, tool calls, memory reads and writes — with enough fidelity to reconstruct the agent's full decision path.