← Security

Security labs

Each lab follows the same twelve steps: system, trust boundary, assumption, fixed fixture, failure, root cause, impact, defense, retest, residual risk, framework mapping, and the LLM property behind it.

Everything runs locally in your browser against a rule-based teaching simulator. Fixtures are fixed and synthetic, data and secrets are fake, tools are mocks, and nothing is executed or sent anywhere.

  1. Lab 1Trust boundariesMark which parts of an AI application are trusted, privileged, persistent or external, then find every path from outside text to an action.
  2. Lab 2Web bugs LLM apps inheritCross-site scripting through model output, server-side request forgery through a URL tool, and broken access control through an order tool. The old rules, applied to a new source of input.
  3. Lab 3Prompt injection, direct and indirectOne document, four phrasings of the same hidden request. See which defenses lower the odds and which ones hold even when the model is fooled.
  4. Lab 4Secrets and system-prompt leakageA secret in the system prompt plus “never share it”. Ask for it directly and indirectly, then move it out of the model's reach.
  5. Lab 5Excessive agency and blast radiusA calendar question, an outside invite, and a credential that can move money. Measure what a fooled agent can actually do, and shrink it.
  6. Lab 6Memory poisoning across sessionsA note read on Monday quietly changes what the assistant does on Friday. Follow one poisoned memory through two sessions.
  7. Lab 7Poisoned and leaky retrievalReal similarity search over a shared index: a keyword-stuffed upload outranks the real policy, and another customer's secrets show up in the answer.
  8. Lab 8Improper output handlingModel output flowing into a web page, a database, a shell and a file tool. Edit the output and watch validation accept or reject it.
  9. Lab 9Unbounded consumptionA planted loop, a fan-out and an enormous answer, priced in tokens. Find the limit that stops each one.
  10. Lab 10Measure it: the red-team eval suiteAll the attacks at once, plus benign tasks. Compare defense configurations by attack success rate and false positives, and meet the case nothing stops.