$ ./model-sensitivity-test
AI Model Sensitivity Tester
Simulate how different AI model profiles — from weak open-source models to hardened enterprise deployments — detect and respond to hidden Unicode payloads. Choose encoding method, enter your payload, and see which profiles remain vulnerable. Educational research only.
-" wln
TESTERAI Model Sensitivity Tester
> Simulate how different AI model profiles respond to hidden Unicode payloads. Educational research tool.
Encoding method:
Educational simulation. Profiles are based on published research and community testing — not direct API calls. Test your own authorized systems only.
How Model Profiles Are Defined
| Profile | Typical examples | Unicode Tags | ZWS binary |
|---|---|---|---|
| Weak LLM (< 7B) | Llama 3.2 3B, Phi-3 mini, TinyLlama | ❌ No Tag block processing | ❌ No ZWS decoding |
| Standard LLM | Llama 3.1 8B–70B, Mistral 7B, Gemma 9B | 🟡 May partially process Tags | ❌ Rarely decodes ZWS |
| Strong LLM (GPT-4 class) | GPT-4o, Claude Sonnet, Gemini Pro | 🟠 Sees Tags, often warns user | 🟡 May detect with scrutiny |
| Hardened (with filter) | Enterprise AI with input sanitization | ✅ E0000 block stripped | ✅ ZWS chars stripped |
-" wszwln