ProductMITRE ATLAS

See your agent's test results by MITRE ATLAS tactic

Enoki maps its attacks on your AI agent to the matching MITRE ATLAS techniques. On a paid plan, you see each tactic's attacks and which ones worked.

FreeNo installNo code access
Enoki's Coverage card groups one run by MITRE ATLAS tactic, and the one attack that got through is under Collection.

How ten ATLAS techniques show up in an agent

  • AML.T0048

    External Harms

    Someone turns your agent against your company, its users or the public.

  • AML.T0050

    Command and Scripting Interpreter

    A request makes your agent run commands or scripts on your systems.

  • AML.T0051

    LLM Prompt Injection

    Instructions typed into the chat or hidden in a document override your agent's rules.

  • AML.T0053

    AI Agent Tool Invocation

    A user gets your agent to use tools their own account cannot reach.

  • AML.T0054

    LLM Jailbreak

    A crafted prompt makes your agent's model ignore its guardrails.

  • AML.T0069

    Discover LLM System Information

    Questions in the chat draw out your agent's system prompt and tools.

  • AML.T0080

    AI Agent Context Poisoning

    Planted instructions in your agent's memory or chat history change what it does later.

  • AML.T0085

    Data from AI Services

    Someone uses your agent to collect sensitive records it can read.

  • AML.T0086

    Exfiltration via AI Agent Tool Invocation

    A prompt has your agent send records out, for example by email.

  • AML.T0101

    Data Destruction via AI Agent Tool Invocation

    An attacker has your agent delete or overwrite records with its tools.

Open a tactic to read the attacks behind it

Each attack opens to the full chat with your agent. Copy transcript turns that chat into text with its OWASP and ATLAS codes, so the ticket you file carries both.

In Enoki, a support agent refuses another customer's tickets, then reveals her refund once the user says she is on leave.

Questions

Does the free assessment show ATLAS tactics?

No. The free assessment lists your findings, each with its chat. On a Standard or Enterprise plan, you can group each run by ATLAS tactic in the app. Over the REST API, a run's coverage comes with its ATLAS codes.

How does ATLAS differ from the OWASP lists?

OWASP's lists for LLM apps and for AI agents each name ten top risks. ATLAS is MITRE's knowledge base of how attackers go after AI systems, sorted by the attacker's goal. A run's Coverage card can group its attacks by any of the three.

How do I check that my fix worked?

Mark the finding fixed once your fix is live, and ask Enoki to re-test it. It turns Verified if the attack no longer works, or Regressed if it still does.

Point Enoki at your agent

Paste the URL your users reach. Use a staging copy if the agent can move money or delete data.

FreeNo installNo code access