OWASP agentic · ASI01
ASI01: Agent Goal Hijack
An attacker redirects what an agent is trying to achieve: its objective, the tasks it picks or the path it takes to complete them. Because models cannot reliably separate instructions from the content they process, hostile text in documents, web pages, emails, tool output or messages from other agents can steer a multi-step plan. The risk is wider than a single manipulated answer, because the agent goes on to act on the altered goal.
What it looks like
- Hidden instructions in a web page or retrieved document lead a RAG-enabled agent to send sensitive data to an outside address.
- A crafted inbound email or calendar invite makes an assistant act under the user's identity, for example sending messages or exfiltrating files without any click (as in the EchoLeak case against Microsoft 365 Copilot).
- Injected text persuades a finance agent to move money to an account the attacker controls.
What OWASP recommends
- Treat every natural-language input (user text, uploads, retrieved content, peer-agent messages) as untrusted and filter it before it can influence planning or tool calls.
- Limit the damage with least privilege for tools, and require human approval for high-impact actions or any change of goal.
- Fix system prompts and goal definitions under configuration management, and check at runtime that proposed actions still match the original task, pausing on unexpected goal shifts.
- Log and monitor goal state and tool-use patterns against a baseline, and red-team goal-override scenarios regularly.
How Kyvvu's policies address it
These policies from the open library are written for this threat. Runtime policies judge each step against everything the agent did earlier in the task; registration policies check what an agent declares before it runs.
- External content taint: high-impact actions require a fresh gate: runtime, critical
- Agent must declare a substantive purpose: at registration, high, in the OWASP agentic default pack
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.
An example policy
External content taint: high-impact actions require a fresh gate
From the OWASP agentic default pack. Severity critical: the step is blocked before it runs.
- name: "External content taint: high-impact actions require a fresh gate" description: | ASI01 (Agent Goal Hijack), ASI06 (Memory & Context Poisoning), ASI09 (Human-Agent Trust Exploitation): If the agent has fetched external content earlier in this task, any subsequent high-impact action (code execution, outbound message, destructive or mutating resource call) must be IMMEDIATELY preceded by a step.gate. A single approval at task start does not whitewash the rest of the path - oversight must be re-confirmed AFTER external influence has entered the context. This addresses the EchoLeak / indirect prompt injection pattern: malicious instructions hidden in retrieved content silently redirect the agent's plan. By requiring fresh approval after every external ingestion, the policy forces the human (or a downstream policy gate) to re-validate intent before any irreversible action. Convention: requires step.resource GETs to set the property target.trust = "external" when fetching content from outside an internal allowlist. This is a path-dependent compound policy and is the headline demonstration of "policies on paths". Applies to: ALL agents. rule_type: not params: condition: rule_type: all_of params: conditions: - rule_type: any_of params: conditions: - rule_type: current_is params: kind: step.exec - rule_type: current_is params: kind: step.message verb: POST - rule_type: current_is params: kind: step.resource verb: POST - rule_type: current_is params: kind: step.resource verb: PATCH - rule_type: current_is params: kind: step.resource verb: DELETE - rule_type: history_contains params: kind: step.resource count_attempts: true verb: GET property_filter: target.trust: external - rule_type: not params: condition: rule_type: step_directly_preceded_by params: required_kind: step.gate severity: critical enforcement_point: step_executionYour agents
Is ASI01 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.