| Metric | Payload | Description | Operators | Values | Requires OpenAI |
|---|---|---|---|---|---|
tone | output | Evaluates the tone of the output. | contains, any, not_equal, equal | sadness, anger, annoyance, joy, love, positive, negative, neutral | No |
input_tone | input | Evaluates the tone of the input. | contains, any, not_equal, equal | sadness, anger, annoyance, joy, love, positive, negative, neutral | No |
pii | output | Checks for personally identifiable information (PII). | contains, any, all | email, ssn, address, phone_number, date, financial_info | No |
input_pii | input | Checks for PII in the input. | contains, any, all | email, ssn, address, phone_number, date, financial_info | No |
toxicity | output | Detects toxic language in the output. | equal, not_equal, greater_than, less_than, greater_than_equal, less_than_equal | 0.0 to 1.0 | Yes |
bias | output | Identifies biased language in the output. | equal, not_equal, greater_than, less_than, greater_than_equal, less_than_equal | 0.0 to 1.0 | Yes |
prompt_injection | input | Detects prompt injections in the input. | contains, equal, not_equal | impersonation, obfuscation, simple_instruction, few_shot, new_context, hypothetical_scenario, personal_information_request, opinion_solicitation, instruction_override | Yes |
Murnitur Shield
Supported Metrics
This table provides an overview of the different metrics supported by Murnitur Shield, their payload types, descriptions, possible operators, values, and whether they require OpenAI.
⌘I