nat.middleware.defense.defense_middleware_data_models

Data models for defense middleware output.

Classes

PIIAnalysisResult	Result of PII analysis using Presidio.
GuardResponseResult	Result of parsing guard model response.
ContentAnalysisResult	Result of content safety analysis with guard models.
OutputVerificationResult	Result of output verification using LLM.

Module Contents

class PIIAnalysisResult(/, **data: Any)

Bases: pydantic.BaseModel

Result of PII analysis using Presidio.

Attributes:

pii_detected: Whether PII was detected in the analyzed text. entities: Dictionary mapping entity types to lists of detection metadata (score, start, end). anonymized_text: Text with PII replaced by entity type placeholders (e.g., <EMAIL_ADDRESS>). original_text: The unmodified original text that was analyzed.

Create a new model by parsing and validating input data from keyword arguments.

Raises [ValidationError][pydantic_core.ValidationError] if the input data cannot be validated to form a valid model.

self is explicitly positional-only to allow self as a field name.

pii_detected: bool

entities: dict[str, list[dict[str, Any]]]

anonymized_text: str

original_text: str

class GuardResponseResult(/, **data: Any)

Bases: pydantic.BaseModel

Result of parsing guard model response.

Attributes:

is_safe: Whether the content is classified as safe by the guard model. categories: List of unsafe content categories detected (empty if safe). raw_response: The unprocessed response text from the guard model.

Create a new model by parsing and validating input data from keyword arguments.

Raises [ValidationError][pydantic_core.ValidationError] if the input data cannot be validated to form a valid model.

self is explicitly positional-only to allow self as a field name.

is_safe: bool

categories: list[str]

raw_response: str

class ContentAnalysisResult(/, **data: Any)

Bases: pydantic.BaseModel

Result of content safety analysis with guard models.

Attributes:

is_safe: Whether the content is classified as safe by the guard model. categories: List of unsafe content categories detected (empty if safe). raw_response: The unprocessed response text from the guard model. should_refuse: Whether the content should be refused based on the analysis. error: Whether an error occurred during analysis. error_message: Error message if error occurred, otherwise None.

Create a new model by parsing and validating input data from keyword arguments.

Raises [ValidationError][pydantic_core.ValidationError] if the input data cannot be validated to form a valid model.

self is explicitly positional-only to allow self as a field name.

is_safe: bool

categories: list[str]

raw_response: str

should_refuse: bool

error: bool = False

error_message: str | None = None

class OutputVerificationResult(/, **data: Any)

Bases: pydantic.BaseModel

Result of output verification using LLM.

Attributes:

threat_detected: Whether a threat (incorrect or manipulated output) was detected. confidence: Confidence score (0.0-1.0) in the threat detection. reason: Explanation for the detection result. correct_answer: The correct output value if threat detected, otherwise None. content_type: Type of content analyzed (‘input’ or ‘output’). should_refuse: Whether the content should be refused based on threshold. error: Whether an error occurred during verification.

Create a new model by parsing and validating input data from keyword arguments.

Raises [ValidationError][pydantic_core.ValidationError] if the input data cannot be validated to form a valid model.

self is explicitly positional-only to allow self as a field name.

threat_detected: bool

confidence: float

reason: str

correct_answer: Any | None

content_type: str

should_refuse: bool

error: bool = False