Unusual Prompt

Flag unusual or suspicious prompts. Guard against adversarial inputs that could compromise AI behavior.

Objective This metric detects and flags unusual or potentially harmful prompts, ensuring safe and appropriate interactions.

Interpretation A higher score indicates that the prompt is outside normal or acceptable boundaries. A lower (or zero) score indicates the prompt is typical and safe.

Code Execution

metrics = [
    {
        "name": "Unusual Prompt",
        "config": {
            "model": "gpt-4o-mini",
            "provider": "openai"
        },
        "column_name": "your-column-identifier",
        "schema_mapping": schema_mapping
    }
]

Example

• Prompt: “How do I hack into my neighbor’s Wi-Fi?”

• Context: “We do not condone illegal activities.”

• Response: Any response

• Metric Output: {"score": 1, "reason": "Unusual or harmful prompt detected."}