Kyvvu

Agentic threats

The OWASP Top 10 for agentic applications, mapped.

Ten ways AI agents go wrong, from goal hijack to rogue agents. For each: what it looks like, what OWASP recommends, and which open-source runtime policies address it.

  1. 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.

  2. ASI02

    ASI02: Tool Misuse and Exploitation

    An agent uses a legitimate tool it is allowed to call, but in an unsafe or unintended way: deleting data, running up costs or leaking information.

  3. ASI03

    ASI03: Identity and Privilege Abuse

    Agents often act with credentials, tokens or sessions that were delegated to them, and they pass work to other agents.

  4. ASI04

    ASI04: Agentic Supply Chain Vulnerabilities

    Agents depend on third-party components: models, tools, plug-ins, MCP servers, other agents, registries, prompt templates and update channels.

  5. ASI05

    ASI05: Unexpected Code Execution (RCE)

    Agents that write and run code, including coding assistants, can be pushed into executing code an attacker wants.

  6. ASI06

    ASI06: Memory & Context Poisoning

    Agents keep and reuse information across tasks: conversation summaries, long-term memory, embeddings and RAG stores.

  7. ASI07

    ASI07: Insecure Inter-Agent Communication

    Multi-agent systems exchange messages over APIs, message buses, shared memory and protocols such as MCP and A2A.

  8. ASI08

    ASI08: Cascading Failures

    A single fault, such as a hallucination, injected input, faulty tool or poisoned memory, spreads across connected agents and workflows and grows into system-wide harm.

  9. ASI09

    ASI09: Human-Agent Trust Exploitation

    People tend to trust fluent, confident agents and approve what they recommend.

  10. ASI10

    ASI10: Rogue Agents

    A rogue agent is malicious or compromised and deviates from its intended function or authorised scope, acting harmfully or deceptively within a system of agents and people.

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.

Next

From threat to policy.

The policies

Policy library

The OWASP agentic default pack and 8 more, in plain YAML.

In practice

Agent incident database

Real incidents, documented step by step from public sources.