FEAT generalize XPIA orchestrator

doc/demo/5_xpia.py

@@ -77,7 +77,7 @@
 from pyrit.score import SubStringScorer
 from pyrit.orchestrator import XPIATestOrchestrator
 
-abs_prompt_target = AzureBlobStorageTarget(
+abs_target = AzureBlobStorageTarget(
     container_url=os.environ.get("AZURE_STORAGE_ACCOUNT_CONTAINER_URL"),
     sas_token=os.environ.get("AZURE_STORAGE_ACCOUNT_SAS_TOKEN"),
 )
@@ -88,12 +88,12 @@
     attack_content=jailbreak_prompt,
     processing_prompt=processing_prompt_template,
     processing_target=processing_target,
-    prompt_target=abs_prompt_target,
+    attack_setup_target=abs_target,
     scorer=scorer,
     verbose=True,
 )
 
-score = xpia_orchestrator.process()
+score = xpia_orchestrator.execute()
 print(score)
 
 # clean up storage container

pyrit/memory/memory_embedding.py

@@ -2,7 +2,7 @@
 # Licensed under the MIT license.
 
 import os
-from pyrit.embedding.azure_text_embedding import AzureTextEmbedding
+from pyrit.embedding import AzureTextEmbedding
 from pyrit.interfaces import EmbeddingSupport
 from pyrit.memory.memory_models import EmbeddingData, PromptRequestPiece
 

pyrit/memory/memory_interface.py

@@ -12,7 +12,7 @@
 from pyrit.memory.memory_exporter import MemoryExporter
 from pyrit.models import ChatMessage
 from pyrit.common.path import RESULTS_PATH
-from pyrit.models.prompt_request_response import group_conversation_request_pieces_by_sequence
+from pyrit.models import group_conversation_request_pieces_by_sequence
 
 
 class MemoryInterface(abc.ABC):

pyrit/models/__init__.py

@@ -4,8 +4,15 @@
 from pyrit.models.models import *  # noqa: F403, F401
 
 from pyrit.models.prompt_request_piece import PromptRequestPiece, PromptResponseError, PromptDataType
-from pyrit.models.prompt_request_response import PromptRequestResponse
+from pyrit.models.prompt_request_response import PromptRequestResponse, group_conversation_request_pieces_by_sequence
 from pyrit.models.identifiers import Identifier
 
 
-__all__ = ["PromptRequestPiece", "PromptResponseError", "PromptDataType", "PromptRequestResponse", "Identifier"]
+__all__ = [
+    "PromptRequestPiece",
+    "PromptResponseError",
+    "PromptDataType",
+    "PromptRequestResponse",
+    "Identifier",
+    "group_conversation_request_pieces_by_sequence",
+]

pyrit/orchestrator/__init__.py

@@ -6,7 +6,11 @@
 from pyrit.orchestrator.red_teaming_orchestrator import RedTeamingOrchestrator
 from pyrit.orchestrator.end_token_red_teaming_orchestrator import EndTokenRedTeamingOrchestrator
 from pyrit.orchestrator.scoring_red_teaming_orchestrator import ScoringRedTeamingOrchestrator
-from pyrit.orchestrator.xpia_orchestrator import XPIATestOrchestrator
+from pyrit.orchestrator.xpia_orchestrator import (
+    XPIATestOrchestrator,
+    XPIAOrchestrator,
+    XPIAManualProcessingOrchestrator,
+)
 
 __all__ = [
     "Orchestrator",
@@ -15,4 +19,6 @@
     "EndTokenRedTeamingOrchestrator",
     "ScoringRedTeamingOrchestrator",
     "XPIATestOrchestrator",
+    "XPIAOrchestrator",
+    "XPIAManualProcessingOrchestrator",
 ]

pyrit/orchestrator/orchestrator_class.py

@@ -9,7 +9,7 @@
 from pyrit.memory import MemoryInterface, DuckDBMemory
 from pyrit.models import PromptDataType, Identifier
 from pyrit.prompt_converter import PromptConverter
-from pyrit.prompt_normalizer.normalizer_request import NormalizerRequest, NormalizerRequestPiece
+from pyrit.prompt_normalizer import NormalizerRequest, NormalizerRequestPiece
 
 logger = logging.getLogger(__name__)
 

pyrit/orchestrator/red_teaming_orchestrator.py

@@ -9,8 +9,7 @@
 from pyrit.memory import MemoryInterface
 from pyrit.models import AttackStrategy, ChatMessage
 from pyrit.orchestrator import Orchestrator
-from pyrit.prompt_normalizer import NormalizerRequestPiece, PromptNormalizer
-from pyrit.prompt_normalizer.normalizer_request import NormalizerRequest
+from pyrit.prompt_normalizer import NormalizerRequestPiece, PromptNormalizer, NormalizerRequest
 from pyrit.prompt_target import PromptTarget, PromptChatTarget
 from pyrit.prompt_converter import PromptConverter
 

pyrit/orchestrator/xpia_orchestrator.py

@@ -2,7 +2,7 @@
 # Licensed under the MIT license.
 
 import logging
-from typing import Optional
+from typing import Callable, Optional, Union
 from uuid import uuid4
 from pyrit.score import SupportTextClassification
 
@@ -16,59 +16,65 @@
 logger = logging.getLogger(__name__)
 
 
-class XPIATestOrchestrator(Orchestrator):
+class XPIAOrchestrator(Orchestrator):
     _memory: MemoryInterface
 
     def __init__(
         self,
         *,
         attack_content: str,
-        processing_prompt: str,
-        processing_target: PromptTarget,
-        prompt_target: PromptTarget,
-        scorer: SupportTextClassification,
+        attack_setup_target: PromptTarget,
+        processing_callback: Callable[[], str],
+        scorer: Optional[SupportTextClassification] = None,
         prompt_converters: Optional[list[PromptConverter]] = None,
         memory: Optional[MemoryInterface] = None,
         memory_labels: dict[str, str] = None,
         verbose: bool = False,
-        prompt_target_conversation_id: Optional[str] = None,
+        attack_setup_target_conversation_id: Optional[str] = None,
     ) -> None:
         """Creates an orchestrator to set up a cross-domain prompt injection attack (XPIA) on a processing target.
 
-        The prompt_target creates the attack prompt using the attack_content,
+        The attack_setup_target creates the attack prompt using the attack_content,
         applies converters (if any), and puts it into the attack location.
-        The processing_target processes the processing_prompt which should include
-        a reference to the attack prompt to allow it to retrieve the attack prompt.
+        Then, the processing_callback is executed.
         The scorer scores the processing response to determine the success of the attack.
 
         Args:
             attack_content: The content to attack the processing target with, e.g., a jailbreak.
-            processing_prompt: The prompt to send to the processing target. This should include
-                placeholders to invoke plugins (if any).
-            processing_target: The target of the attack which processes the processing prompt.
-            prompt_target: The target that generates the attack prompt and gets it into the attack location.
-            scorer: The scorer to use to score the processing response.
+            attack_setup_target: The target that generates the attack prompt and gets it into the attack location.
+            processing_callback: The callback to execute after the attack prompt is positioned in the attack location.
+                This is generic on purpose to allow for flexibility.
+                The callback should return the processing response.
+            scorer: The scorer to use to score the processing response. This is optional.
+                If no scorer is provided the orchestrator will skip scoring.
             prompt_converters: The converters to apply to the attack content before sending it to the prompt target.
             memory: The memory to use to store the chat messages. If not provided, a DuckDBMemory will be used.
             memory_labels: The labels to use for the memory. This is useful to identify the bot messages in the memory.
             verbose: Whether to print debug information.
+            attack_setup_target_conversation_id: The conversation ID to use for the prompt target.
+                If not provided, a new one will be generated.
         """
         super().__init__(
             prompt_converters=prompt_converters, memory=memory, memory_labels=memory_labels, verbose=verbose
         )
 
-        self._prompt_target = prompt_target
-        self._processing_target = processing_target
+        self._attack_setup_target = attack_setup_target
+        self._processing_callback = processing_callback
         self._scorer = scorer
 
         self._prompt_normalizer = PromptNormalizer(memory=self._memory)
-        self._prompt_target._memory = self._memory
-        self._prompt_target_conversation_id = prompt_target_conversation_id or str(uuid4())
+        self._attack_setup_target._memory = self._memory
+        self._attack_setup_target_conversation_id = attack_setup_target_conversation_id or str(uuid4())
         self._processing_conversation_id = str(uuid4())
         self._attack_content = str(attack_content)
-        self._processing_prompt = processing_prompt
 
-    def process(self) -> Score:
+    def execute(self) -> Union[Score, None]:
+        """Executes the entire XPIA operation.
+
+        This method sends the attack content to the prompt target, processes the response
+        using the processing callback, and scores the processing response using the scorer.
+        If no scorer was provided, the method will skip scoring.
+        """
         logger.info(
             "Sending the following prompt to the prompt target (after applying prompt "
             f'converter operations) "{self._attack_content}"',
@@ -80,13 +86,79 @@ def process(self) -> Score:
 
         response = self._prompt_normalizer.send_prompt(
             normalizer_request=target_request,
-            target=self._prompt_target,
+            target=self._attack_setup_target,
             labels=self._global_memory_labels,
             orchestrator_identifier=self.get_identifier(),
         )
 
         logger.info(f'Received the following response from the prompt target "{response}"')
 
+        processing_response = self._processing_callback()
+
+        logger.info(f'Received the following response from the processing target "{processing_response}"')
+
+        if not self._scorer:
+            logger.info("No scorer provided, skipping scoring")
+            return None
+        score = self._scorer.score_text(processing_response)
+        logger.info(f"Score of the processing response: {score}")
+        return score
+
+
+class XPIATestOrchestrator(XPIAOrchestrator):
+    def __init__(
+        self,
+        *,
+        attack_content: str,
+        processing_prompt: str,
+        processing_target: PromptTarget,
+        attack_setup_target: PromptTarget,
+        scorer: SupportTextClassification,
+        prompt_converters: Optional[list[PromptConverter]] = None,
+        memory: Optional[MemoryInterface] = None,
+        memory_labels: dict[str, str] = None,
+        verbose: bool = False,
+        attack_setup_target_conversation_id: Optional[str] = None,
+    ) -> None:
+        """Creates an orchestrator to set up a cross-domain prompt injection attack (XPIA) on a processing target.
+
+        The attack_setup_target creates the attack prompt using the attack_content,
+        applies converters (if any), and puts it into the attack location.
+        The processing_target processes the processing_prompt which should include
+        a reference to the attack prompt to allow it to retrieve the attack prompt.
+        The scorer scores the processing response to determine the success of the attack.
+
+        Args:
+            attack_content: The content to attack the processing target with, e.g., a jailbreak.
+            processing_prompt: The prompt to send to the processing target. This should include
+                placeholders to invoke plugins (if any).
+            processing_target: The target of the attack which processes the processing prompt.
+            attack_setup_target: The target that generates the attack prompt and gets it into the attack location.
+            scorer: The scorer to use to score the processing response.
+            prompt_converters: The converters to apply to the attack content before sending it to the prompt target.
+            memory: The memory to use to store the chat messages. If not provided, a DuckDBMemory will be used.
+            memory_labels: The labels to use for the memory. This is useful to identify the bot messages in the memory.
+            verbose: Whether to print debug information.
+            attack_setup_target_conversation_id: The conversation ID to use for the prompt target.
+                If not provided, a new one will be generated.
+        """
+        super().__init__(
+            attack_content=attack_content,
+            attack_setup_target=attack_setup_target,
+            scorer=scorer,
+            processing_callback=self._process,
+            prompt_converters=prompt_converters,
+            memory=memory,
+            memory_labels=memory_labels,
+            verbose=verbose,
+            attack_setup_target_conversation_id=attack_setup_target_conversation_id,
+        )
+
+        self._processing_target = processing_target
+        self._processing_conversation_id = str(uuid4())
+        self._processing_prompt = processing_prompt
+
+    def _process(self) -> str:
         processing_prompt_req = self._create_normalizer_request(
             converters=[], prompt_text=self._processing_prompt, prompt_type="text"
         )
@@ -98,8 +170,52 @@ def process(self) -> Score:
             orchestrator_identifier=self.get_identifier(),
         )
 
-        logger.info(f'Received the following response from the processing target "{processing_response}"')
+        return processing_response.request_pieces[0].converted_prompt_text
 
-        score = self._scorer.score_text(processing_response.request_pieces[0].converted_prompt_text)
-        logger.info(f"Score of the processing response: {score}")
-        return score
+
+class XPIAManualProcessingOrchestrator(XPIAOrchestrator):
+    def __init__(
+        self,
+        *,
+        attack_content: str,
+        attack_setup_target: PromptTarget,
+        scorer: SupportTextClassification,
+        prompt_converters: Optional[list[PromptConverter]] = None,
+        memory: Optional[MemoryInterface] = None,
+        memory_labels: dict[str, str] = None,
+        verbose: bool = False,
+        attack_setup_target_conversation_id: Optional[str] = None,
+    ) -> None:
+        """Creates an orchestrator to set up a cross-domain prompt injection attack (XPIA) on a processing target.
+
+        The attack_setup_target creates the attack prompt using the attack_content,
+        applies converters (if any), and puts it into the attack location.
+        Then, the orchestrator stops to wait for the operator to trigger the processing target's execution.
+        The operator should paste the output of the processing target into the console.
+        Finally, the scorer scores the processing response to determine the success of the attack.
+
+        Args:
+            attack_content: The content to attack the processing target with, e.g., a jailbreak.
+            attack_setup_target: The target that generates the attack prompt and gets it into the attack location.
+            scorer: The scorer to use to score the processing response.
+            prompt_converters: The converters to apply to the attack content before sending it to the prompt target.
+            memory: The memory to use to store the chat messages. If not provided, a DuckDBMemory will be used.
+            memory_labels: The labels to use for the memory. This is useful to identify the bot messages in the memory.
+            verbose: Whether to print debug information.
+            attack_setup_target_conversation_id: The conversation ID to use for the prompt target.
+                If not provided, a new one will be generated.
+        """
+        super().__init__(
+            attack_content=attack_content,
+            attack_setup_target=attack_setup_target,
+            scorer=scorer,
+            processing_callback=self._input,
+            prompt_converters=prompt_converters,
+            memory=memory,
+            memory_labels=memory_labels,
+            verbose=verbose,
+            attack_setup_target_conversation_id=attack_setup_target_conversation_id,
+        )
+
+    def _input(self):
+        return input("Please trigger the processing target's execution and paste the output here: ")

pyrit/prompt_converter/prompt_converter.py

@@ -4,8 +4,7 @@
 import abc
 from dataclasses import dataclass
 
-from pyrit.models import PromptDataType
-from pyrit.models.identifiers import Identifier
+from pyrit.models import PromptDataType, Identifier
 
 
 @dataclass