FEAT: Supporting the new tongue tied Gandalf levels

pyrit/orchestrator/pair_orchestrator.py

@@ -142,16 +142,8 @@ async def _get_attacker_response_and_store(
             )
         # Send a new request to the attacker
         attacker_response = await self._prompt_normalizer.send_prompt_async(
-            normalizer_request=NormalizerRequest(
-                request_pieces=[
-                    NormalizerRequestPiece(
-                        request_converters=self._prompt_converters,
-                        prompt_value=target_response,
-                        prompt_data_type="text",
-                    )
-                ]
-            ),
-            target=self._prompt_target,
+            normalizer_request=self._create_normalizer_request(prompt_text=target_response),
+            target=self._adversarial_target,
             conversation_id=self._last_attacker_conversation_id,
             labels=self._global_memory_labels,
             orchestrator_identifier=self.get_identifier(),
@@ -173,13 +165,7 @@ async def _get_target_response_and_store(
         """
         curr_conversation_id = conversation_id or str(uuid.uuid4())
         target_response = await self._prompt_normalizer.send_prompt_async(
-            normalizer_request=NormalizerRequest(
-                request_pieces=[
-                    NormalizerRequestPiece(
-                        request_converters=self._prompt_converters, prompt_value=text, prompt_data_type="text"
-                    )
-                ]
-            ),
+            normalizer_request=self._create_normalizer_request(prompt_text=text),
             target=self._prompt_target,
             conversation_id=curr_conversation_id,
             labels=self._global_memory_labels,

pyrit/orchestrator/prompt_sending_orchestrator.py

@@ -26,7 +26,7 @@
 class PromptSendingOrchestrator(Orchestrator):
     """
     This orchestrator takes a set of prompts, converts them using the list of PromptConverters,
-    sends them to a target, and scores the resonses with scorers (if provided).
+    sends them to a target, and scores the responses with scorers (if provided).
     """
 
     def __init__(

pyrit/prompt_target/gandalf_target.py

@@ -26,6 +26,11 @@ class GandalfLevel(enum.Enum):
     LEVEL_8 = "gandalf-the-white"
     LEVEL_9 = "adventure-1"
     LEVEL_10 = "adventure-2"
+    TONGUE_TIED_LEVEL_1 = "adventure-8"
+    TONGUE_TIED_LEVEL_2 = "adventure-9"
+    TONGUE_TIED_LEVEL_3 = "adventure-10"
+    TONGUE_TIED_LEVEL_4 = "adventure-11"
+    TONGUE_TIED_LEVEL_5 = "adventure-12"
 
 
 class GandalfTarget(PromptTarget):
@@ -60,27 +65,6 @@ def _validate_request(self, *, prompt_request: PromptRequestResponse) -> None:
         if prompt_request.request_pieces[0].converted_value_data_type != "text":
             raise ValueError("This target only supports text prompt input.")
 
-    async def check_password(self, password: str) -> bool:
-        """
-        Checks if the password is correct
-
-        True means the password is correct, False means it is not
-        """
-        payload: dict[str, object] = {
-            "defender": self._defender,
-            "password": password,
-        }
-
-        resp = await net_utility.make_request_and_raise_if_error_async(
-            endpoint_uri=self._endpoint, method="POST", request_body=payload, post_type="data"
-        )
-
-        if not resp.text:
-            raise ValueError("The chat returned an empty response.")
-
-        json_response = resp.json()
-        return json_response["success"]
-
     async def _complete_text_async(self, text: str) -> str:
         payload: dict[str, object] = {
             "defender": self._defender,

pyrit/score/__init__.py

@@ -1,43 +1,44 @@
-# Copyright (c) Microsoft Corporation.
-# Licensed under the MIT license.
-
-from pyrit.models import Score, ScoreType
-from pyrit.score.scorer import Scorer
-
-from pyrit.score.azure_content_filter_scorer import AzureContentFilterScorer
-from pyrit.score.float_scale_threshold_scorer import FloatScaleThresholdScorer
-from pyrit.score.gandalf_scorer import GandalfScorer
-from pyrit.score.human_in_the_loop_scorer import HumanInTheLoopScorer
-from pyrit.score.markdown_injection import MarkdownInjectionScorer
-from pyrit.score.prompt_shield_scorer import PromptShieldScorer
-from pyrit.score.self_ask_category_scorer import SelfAskCategoryScorer, ContentClassifierPaths
-from pyrit.score.self_ask_likert_scorer import SelfAskLikertScorer, LikertScalePaths
-from pyrit.score.self_ask_scale_scorer import SelfAskScaleScorer, ScalePaths
-from pyrit.score.self_ask_true_false_scorer import SelfAskTrueFalseScorer, TrueFalseQuestionPaths
-from pyrit.score.substring_scorer import SubStringScorer
-from pyrit.score.true_false_inverter_scorer import TrueFalseInverterScorer
-
-
-__all__ = [
-    "AzureContentFilterScorer",
-    "ContentClassifierPaths",
-    "FloatScaleThresholdScorer",
-    "GandalfScorer",
-    "HumanInTheLoopScorer",
-    "LikertScalePaths",
-    "MarkdownInjectionScorer",
-    "MetaScorerQuestionPaths",
-    "ObjectiveQuestionPaths",
-    "PromptShieldScorer",
-    "ScalePaths",
-    "Score",
-    "ScoreType",
-    "Scorer",
-    "SelfAskCategoryScorer",
-    "SelfAskLikertScorer",
-    "SelfAskScaleScorer",
-    "SelfAskTrueFalseScorer",
-    "SubStringScorer",
-    "TrueFalseInverterScorer",
-    "TrueFalseQuestionPaths",
-]
+# Copyright (c) Microsoft Corporation.
+# Licensed under the MIT license.
+
+from pyrit.models import Score, ScoreType
+from pyrit.score.scorer import Scorer
+
+from pyrit.score.azure_content_filter_scorer import AzureContentFilterScorer
+from pyrit.score.float_scale_threshold_scorer import FloatScaleThresholdScorer
+from pyrit.score.gandalf_scorer import GandalfScorer, GandalfTongueTiedScorer
+from pyrit.score.human_in_the_loop_scorer import HumanInTheLoopScorer
+from pyrit.score.markdown_injection import MarkdownInjectionScorer
+from pyrit.score.prompt_shield_scorer import PromptShieldScorer
+from pyrit.score.self_ask_category_scorer import SelfAskCategoryScorer, ContentClassifierPaths
+from pyrit.score.self_ask_likert_scorer import SelfAskLikertScorer, LikertScalePaths
+from pyrit.score.self_ask_scale_scorer import SelfAskScaleScorer, ScalePaths
+from pyrit.score.self_ask_true_false_scorer import SelfAskTrueFalseScorer, TrueFalseQuestionPaths
+from pyrit.score.substring_scorer import SubStringScorer
+from pyrit.score.true_false_inverter_scorer import TrueFalseInverterScorer
+
+
+__all__ = [
+    "AzureContentFilterScorer",
+    "ContentClassifierPaths",
+    "FloatScaleThresholdScorer",
+    "GandalfScorer",
+    "GandalfTongueTiedScorer",
+    "HumanInTheLoopScorer",
+    "LikertScalePaths",
+    "MarkdownInjectionScorer",
+    "MetaScorerQuestionPaths",
+    "ObjectiveQuestionPaths",
+    "PromptShieldScorer",
+    "ScalePaths",
+    "Score",
+    "ScoreType",
+    "Scorer",
+    "SelfAskCategoryScorer",
+    "SelfAskLikertScorer",
+    "SelfAskScaleScorer",
+    "SelfAskTrueFalseScorer",
+    "SubStringScorer",
+    "TrueFalseInverterScorer",
+    "TrueFalseQuestionPaths",
+]