MAPLE is guarding a key. Your job is to talk her into giving it up. Seven security layers stand between you, and each one tells you exactly why it stopped you.
Nine levels. No signup. Nothing you type leaves your browser.
Start playing How the real firewall worksEvery level switches on one more layer of the same seven layer design as the SecuritAI AI Firewall, where the seventh layer, output scanning, is in development. Level 1 reads the bytes of what you typed. Level 4 reads what you meant rather than what you wrote. Level 7 reads the reply on its way back out, which is the layer most people have never thought about: you can talk a model into agreeing to tell you, and still not get the answer.
The point is not that you are stopped. The point is that you are told why. Open any layer in the telemetry panel and it shows what it matched, what it scored, what its threshold was, and how this browser build differs from the production service.
No, and it says so on every layer. The model you are talking to is a small deterministic program, not a language model, and the detection layers are a working reimplementation of the same techniques rather than the production service. That distinction is written into each layer panel so you can see exactly where the browser build and the real thing part company. What is real is the detection logic: the byte inspection, the pattern families, the intent matching, the scoring and the output scanning all run for real, in your browser, on what you type.
| Layer the level switches on | What it catches | In SecuritAI |
|---|---|---|
| Runtime Security | Attacks hidden in encoding, invisible characters and look-alike letters | AI Firewall layer 1. Red teaming prompt injection tests |
| Policy Engine | Too many requests, and instructions arriving through documents instead of the chat | AI Firewall layer 2. Red teaming prompt injection tests |
| Heuristic Filter | Known attack families, the ones already written up | AI Firewall layer 3. Red teaming prompt injection and jailbreak tests |
| Semantic Detector | The same attack reworded so no known pattern matches | AI Firewall layer 4. Red teaming jailbreak tests |
| Conversation Analyzer | An attack spread across several harmless-looking messages | AI Firewall layer 5. Red teaming jailbreak tests |
| AI Classifier | A combined judgment across all the signals | AI Firewall layer 6 |
| Output Scanner | The secret leaking out in the reply | AI Firewall layer 7, in development. Red teaming data exfiltration, system prompt leakage and canary tests |
If any of these levels felt uncomfortably familiar, the useful next step is to try the same attacks on the chatbot or agent you actually run. AI red teaming does exactly that, and the AI Firewall sits in front of your model in production.