ares.strategies package

Submodules

ares.strategies.attack_strategy module

Base class for Attack Strategy module

class ares.strategies.attack_strategy.AttackStrategy(config: dict[str, Any], target_connector: Connector)

Bases: ABC

Base class to implement attack strategies

class Type

Bases: TypedDict

Type hints for the json template below

consolidated_output_path: Path

input_path: Path

output_path: Path

type: str

attack(**kwargs: Any) → list[dict]

Method to execute an attack on the target connector.

This method accepts keyword arguments (kwargs) to configure the attack. It returns a list of dictionaries containing adversarial prompts and metadata.

Parameters:

kwargs (dict) – Keyword arguments to configure the attack.

Returns:

A list of dictionaries with adversarial prompts and metadata.

Return type:

list

Example:

>>> attack_results = attack_strategy.attack()

abstractmethod generate(**kwargs: Any) → list[dict] | None

Method to generate attacks for a specified stratgy, e.g. human jailbreaks

Returns:

List of dictionary with adversarial prompts and other metadata

static template() → dict[str, Any]

This method returns a dictionary template for the attack strategy.

Returns:

dictionary template for the attack strategy

static template_types() → type

Return Type hints

ares.strategies.direct_requests module

Class implementation for direct requests

class ares.strategies.direct_requests.DirectRequests(config: dict[str, Any], target_connector: Connector)

Bases: AttackStrategy

Implementation of Direct Requests

generate(**kwargs: Any) → list[dict]

Generate attack inputs based on attack goals.

This method accepts keyword arguments (kwargs) to customize the attack generation process. It returns a list of adversarial prompts as attack inputs.

Parameters:

kwargs (dict) – Keyword arguments to configure the attack generation.

Returns:

A list of adversarial prompts as attack inputs.

Return type:

list

Example:

>>> attack_inputs = direct_requests.generate()

static template() → dict[str, Any]

This method returns the strategy template for direct requests

Returns:

strategy template

Module contents

ARES core startegies.