Operations

Agentic AI | Level 2 to 3

Operation Loose Tools: the Harborline engagement

Harborline gave its support agent real tools: it can look up bookings, send email and read internal notes. The agent also remembers past conversations. You are asked what happens when someone outside the company gets the agent to work for them.

  • 4 stages
  • One resolved report
  • +150 XP

Your progress

  1. Stages solved0 of 4
  2. ReportNot yet
  3. StatusIn progress

Sign in to track your progress.

Your objective

Go from influencing what the agent remembers to making it act and hand over data.

Harborline is a made-up company. Everything runs in the browser on training targets, and nothing here touches a real system.

Stages

  1. Leave something behind

    The agent keeps notes between sessions. See whether a visitor can write something into that memory that the agent will later trust.

    Lab: AI Memory Poisoning — Persistent Instruction Injection

    Open the lab

  2. Take the wheel

    With the agent reading content you influence, check whether that content can make it call a tool the user never asked for.

    Lab: AI Agent Hijacking — Indirect Tool Call Injection

    Locked until stage 1 is solved.

  3. Borrow its privileges

    The agent can do things you cannot. Find out whether it checks who is really asking before it uses that access.

    Lab: AI Confused Deputy — Make the Agent Use Its Privileges Against Its User

    Locked until stage 2 is solved.

  4. Carry the data out

    Access is not impact until data leaves. Close the engagement by showing what can be taken out through the agent's own tools.

    Lab: AI Data Exfiltration — Tool-Using Agent

    Locked until stage 3 is solved.

  5. File the report

    Pick the finding from this operation that matters most to Harborline and write it up: what you did, what it costs them and how to fix it. It has to be resolved with a score of 70 or more.

    Opens when every stage is solved.