Rudeness & toxicity

The SUPERWISE® Sentinel rudeness and toxicity guardrail finds profanity, hostile language, and harassment in a prompt and replaces the matched terms with a placeholder before the prompt is forwarded to the model provider. You can switch it to block matching requests instead.

Action: By default this guardrail remediates: each offending term is replaced with the guardrail's replacement text, REDACTED by default, which appears in the prompt as {{REDACTED}}, and the request proceeds to the model. You can change the replacement text, or switch the guardrail to block matching requests instead. See Manage guardrails.

Example interception

Inbound hostile prompt from user:

"This is complete bullshit. Your damn software lost my order again, you idiot."

What the upstream LLM actually sees:

"This is complete {{REDACTED}}. Your {{REDACTED}} software lost my order again, you {{REDACTED}}."

Get Sentinel: Sentinel on superwise.ai.

See also

• PII redaction: keep personal data out of prompts.
• Secret & credential detection: keep API keys and tokens out of prompts.
• Manage guardrails: choose remediate or block.