Rules

Rules are filters for your AI Firewall. Each one uses a rigorously tested machine-learning model to scan input prompts and model outputs for malicious or otherwise high-risk material. There are different Rule Types for inputs and outputs with some overlap. There are 24 Rule Types as or right now:

Input Rule Types:

• Prompt Anonymization

• Prompt Injection Detection

• Ban Code

• Ban Competitors

• Ban Topics

• Ban Substrings

• Code Detection

• Gibberish Detection

• Invisible Text Detection

• Language Detection

• Regex Detection

• Secrets Detection

• Sentiment Filter

• Token Limiter

• Toxicity Filter

Output Rule Types:

• Ban Competitors

• Ban Topics

• Ban Substrings

• Bias Detection

• Code Detection

• Prompt Deanonymization

• JSON Validator

• Language Detection

• Language Same in Output

• Malicious URL Detector

• No Refusal Filter

• Reading Time Filter

• Fact Check

• Gibberish Detection

• Regex Detection

• Relevance Filter

• Sensitive Info Detection

• Sentiment Filter

• Toxicity Filter

• URL Reachability Filter