Nice To E-Meet You!



    What marketing services do you need for your project?

    OWASP Top 10 For LLM And Agentic Apps Explained (2027)

    Last updated: November 16, 2026

    The OWASP Top 10 for LLM Applications lists the most important security risks in apps built on large language models, led by prompt injection, sensitive information disclosure and supply chain weaknesses. OWASP’s newer Top 10 for Agentic Applications covers risks that appear when AI agents plan and act with tools, such as goal hijacking, tool misuse and identity abuse. Together they are the most widely used checklists for securing AI apps in 2027.

    Below we explain each risk in plain language with the controls that help most. Always check the official OWASP GenAI Security Project pages for the latest wording, since the lists are updated as threats change.

    OWASP Top 10 For LLM Applications

    IDRiskWhat it meansKey controls
    LLM01Prompt InjectionInputs or hidden content override the model’s instructionsTreat content as untrusted, least privilege, detection tools
    LLM02Sensitive Information DisclosureThe model reveals private or confidential dataData minimization, output filtering, access control
    LLM03Supply ChainRisky models, datasets or pluginsVet and scan models, pin versions
    LLM04Data And Model PoisoningTampered training or fine-tuning dataData provenance and validation
    LLM05Improper Output HandlingModel output used unsafely downstreamValidate and encode outputs
    LLM06Excessive AgencyThe model can do too much through toolsLimit tools, scopes and require approvals
    LLM07System Prompt LeakageSecrets or logic exposed via the system promptKeep secrets out of prompts
    LLM08Vector And Embedding WeaknessesAttacks on RAG stores and embeddingsAccess control on vector data, input checks
    LLM09MisinformationConfident but wrong outputGrounding, citations, human review
    LLM10Unbounded ConsumptionAbuse that drives up cost or causes outagesRate limits, quotas, monitoring

    OWASP Top 10 For Agentic Applications

    The agentic list focuses on what changes when AI can plan, remember and act. Its themes include:

    • Agent goal hijacking, where injected content changes what the agent is trying to do.
    • Tool misuse and exploitation, where legitimate tools are used in harmful ways.
    • Identity and privilege abuse, where agents act with more access than they should.
    • Agentic supply chain risks from third-party tools, MCP servers and plugins.
    • Unexpected code execution by agents that can run code.
    • Memory and context poisoning that persists across sessions.
    • Insecure communication between agents.
    • Cascading failures across multi-agent systems.
    • Human-agent trust exploitation, where people over-trust agent output.
    • Rogue agents acting outside their intended scope.

    Practical Controls That Cover Most Risks

    • Least privilege for every tool, token and data source.
    • Human approval for high-impact actions.
    • Input and output inspection, for example with prompt injection detection tools.
    • Isolation and sandboxing for code execution.
    • Full logging and monitoring of agent actions.
    • Regular red teaming; see top AI security consulting firms.

    For a step-by-step version, use our AI agent security checklist.

    Frequently Asked Questions

    What is the OWASP Top 10 for LLM applications?

    It is a community-built list of the ten most important security risks in applications that use large language models, maintained by the OWASP GenAI Security Project.

    What is the number one LLM security risk?

    Prompt injection is ranked first, because inputs or hidden content can override a model’s instructions.

    Is there an OWASP list for AI agents?

    Yes. OWASP published a Top 10 for Agentic Applications focused on risks such as goal hijacking, tool misuse and identity abuse.

    How do I use the OWASP lists?

    Use them as a checklist during design, testing and procurement, and ask vendors and testers to map their controls and findings to them.

    Related Reading

    AI Security And Cyber Threats hub, How to secure AI agents and MCP integrations, Top MCP security tools for 2027, AI red teaming vs penetration testing.

      Once a week you will get the latest articles delivered right to your inbox