Validates that the output does not contain banned words, using fuzzy search.

Validates that the input string does not contain toxic language based on a BERT model.

Validates that the text is free from biases related to age, gender, sex, ethnicity, religion, etc.

Flags mentions of competitors. Fixes responses by filtering out competitor names.

A Guardrails AI validator to check if the LLM-generated text contains a substring.

Validates that the input string matches a specified cucumber expression.

Detects attempts to circumvent safeguards in model conditioning.

Detects personally identifiable information (PII) in text, using Microsoft Presidio.

Finds prompt injection using the Rebuff prompt library.

Detects system prompt leakage using fuzzy string matching (via `rapidfuzz`) to compare the LLM's response against a provided system prompt.

Checks if an endpoint can be reached by making a request to it.

Check if a string or list ends with a specified string or list.

This rule checks for the use of particular SQL predicates in the query. It is important to exclude SQL predicates from the query to prevent SQL injection attacks.

This validator checks if the extracted summary sentences match the original document.

A Guardrails AI validator to detect gibberish text.

Detects personally identifiable information (PII) in text.

Ensure content contains a url.

Identifies internal domains in a string output.

A llama based validator which checks whether a given prompt is safe/unsafe by specifying a set of policies and lists the violating policies when applicable.

Grade the generated response based on provided criteria.

Validates logical consistency and detects logical fallacies in the model output. Attempts to correct logical fallacies if found.

Passes when totally lowercase.

Validates that the generated text does not contain any drug names.

A Guardrails AI validator to detect NSFW text.

This validator checks if the input is a single line of text.

Ensure generated output is polite.

Detects personally identifiable information (PII) in text.

Checks for profanity in text, using the alt-profanity-check library.

A Guardrails validator that scores prompts for injection attempts via a secondary LLM.

Compares embeddings of generated and source texts to calculate provenance.

A validator for ensuring the factuality and reducing brand risk in generated content.

Detects and removes hallucinations from LLM-generated text using an NLI model to validate it against a provided context.

Makes a second request to the LLM, asking it if its original response was relevant to the prompt.

Validates that the generated text contains a price quote.

Parses text to find its readability as a US grade level number (0-12).

Ensures that any generated text is less than a maximum expected reading time.

Identifies redundant sentences in text using fuzzy matching.

Ensure content matches a provided regular expression. This can be used to validate content such as email addresses, phone numbers, and more.

Validates that the reference text contains information relevant to answering the original question.

Evaluate generated output using a provided question.

Ensure generated output is polite.

Checks that a text stays on a set of valid topics and avoids invalid ones.

Checks if a generated summary covers topics present in a source document.

Detects the secrets present in text by matching against common patterns for API keys and other sensitive information.

A Guardrails AI validator that detects sensitive topics in text.

A Gemma based validator for moderating user prompts to guard against harmful content by specifying a policy.

Checks if some generated text is similar to a provided document.

Checks if a value is similar to a list of previously known correct values.

Validates that the input string does not contain negative statements about Sky Electric.

Identifies and flags toxic language in text to ensure communications remain professional and appropriate.

Detects toxic language in LLM-generated text using an LLM as the detection backbone. Evaluates text across seven toxicity categories: toxicity, severe toxicity, obscene, threat, insult, identity attack, and sexual explicit content.

Passes when value is *exactly* two words.

A Guardrails AI input validator that validates a prompt for unusualness and trickery.

Passes when totally uppercase.

Verifies an LLM-generated address using Google Maps' Address Validation API.

Checks if a given string is a valid choice from a list of choices.

Guardrails validator that checks for HTML parseability.

Ensure content is parseable as valid JSON.

Ensures the length of a string or list falls between a minimum and maximum.

Ensures that a generated output is a valid OpenAPI Specification.

Assess whether a generated number is between a maximum and minimum value.

Validates whether the given SQL code is syntactically correct using. Optionally accepts a database schema to validate against using SQLAlchemy.

Validates that text is a syntactically-valid URL.

Scans LLM outputs for strings that could cause browser script execution downstream.

A Guardrails AI validator that detects and removes hallucinated text based off Wikipedia.