MAINT remove PromptChatTarget by hannahwestra25 · Pull Request #1532 · microsoft/PyRIT

hannahwestra25 · 2026-03-23T18:26:10Z

Description

PromptChatTarget has historically was an interface for a target that supports multi-turn conversations with editable history. This PR decouples those concerns by moving capability signaling to TargetCapabilities making PromptChatTarget an empty deprecated subclass.

Also, deprecate set_system_prompt and directing users to use prepended messages. Targets still set system prompts via an internal method _set_target_system_prompt

Changes

All attack strategies, scenarios, scorers, and converters that previously required a PromptChatTarget now accept any PromptTarget Compatibility is enforced at construction time through TargetCapabilities validate function with supports_multi_turn & supports_editable_history instead of a type check.

Tests and Documentation

[in progress] running all target notebooks & integration tests

…emove_prompt_chat_target

…om/hannahwestra25/PyRIT into hawestra/remove_prompt_chat_target

rlundeen2 · 2026-03-24T18:55:45Z

pyrit/executor/attack/component/conversation_manager.py

+            new_item="AttackParameters.prepended_conversation",
+            removed_in="0.14.0",
+        )
+        target._set_target_system_prompt(


The one thing I want to check is that this flow still works in the few places you can have a system prompt but are not editable_history, like realtime_target.

Because prepended_conversation is essentially how we edit the conversation. But then we also need a way to set the system prompt for that target specifically.

At the same time, we've had these really nasty bugs where attacks will call set_system_prompt on a target and it messes everything up. So I LIKE that we're setting it via conversation history. But we just need to figure out these special cases.

rlundeen2 · 2026-03-24T18:58:54Z

pyrit/executor/attack/multi_turn/crescendo.py

        )

-        self._adversarial_chat.set_system_prompt(
+        self._adversarial_chat._set_target_system_prompt(


this may already be the fix for what I mentioned earlier about realtime. But we want to verify.

rlundeen2 · 2026-03-24T19:00:54Z

pyrit/executor/attack/multi_turn/tree_of_attacks.py

        self._adversarial_chat = attack_adversarial_config.target
-        if not isinstance(self._adversarial_chat, PromptChatTarget):
-            raise ValueError("The adversarial target must be a PromptChatTarget for TAP attack.")
+        self._adversarial_chat.capabilities.validate(


validate centrally for adversarial_chat

We could probably do this in the adversarial_chat_config object? Or a base attack class

rlundeen2 · 2026-03-24T19:02:12Z

pyrit/executor/attack/core/attack_config.py


    # Adversarial chat target for the attack
-    target: PromptChatTarget
+    target: PromptTarget


yeah, this needs to be validated to be multi-turn and editable; then we could potentially take out the validation from TAP

rlundeen2 · 2026-03-24T19:03:26Z

pyrit/executor/attack/multi_turn/multi_prompt_sending.py

            )

+        # Validate that the adversarial chat target supports required capabilities
+        if adversarial_chat:


nit: validate centrally for adversarial_chat

rlundeen2 · 2026-03-24T19:05:21Z

pyrit/executor/promptgen/anecdoctor.py


        # Configure the target with the system prompt
-        self._objective_target.set_system_prompt(
+        self._objective_target._set_target_system_prompt(


likely want to validate _objective_target here

rlundeen2 · 2026-03-24T19:05:43Z

pyrit/executor/attack/multi_turn/crescendo.py

-                "Use RedTeamingAttack or TreeOfAttacksWithPruning instead."
-            )
+        self._objective_target.capabilities.validate(
+            required={"supports_multi_turn"},


also needs editable_history

rlundeen2 · 2026-03-24T19:06:42Z

pyrit/executor/attack/multi_turn/tree_of_attacks.py

@@ -1293,7 +1293,7 @@ def __init__(

        Raises:


objective_target also needs to validate multi_turn and editable history

rlundeen2 · 2026-03-24T19:07:33Z

pyrit/executor/attack/single_turn/context_compliance.py

-        objective_target: PromptChatTarget = REQUIRED_VALUE,  # type: ignore[assignment]
+        objective_target: PromptTarget = REQUIRED_VALUE,  # type: ignore[assignment]
        attack_adversarial_config: AttackAdversarialConfig,
        attack_converter_config: Optional[AttackConverterConfig] = None,


need to validate objective_target supports multi-turn and editable history

rlundeen2 · 2026-03-24T19:07:48Z

pyrit/executor/attack/single_turn/flip_attack.py

        """
        Args:
-            objective_target (PromptChatTarget): The target system to attack.
+            objective_target (PromptTarget): The target system to attack.


need to validate objective_target supports multi-turn and editable history

rlundeen2 · 2026-03-24T19:09:28Z

pyrit/executor/promptgen/fuzzer/fuzzer_expand_converter.py

@@ -56,7 +56,7 @@ async def convert_async(self, *, prompt: str, input_type: PromptDataType = "text

        conversation_id = str(uuid.uuid4())


need to validate objective_target supports setting the system prompt

rlundeen2 · 2026-03-24T19:09:51Z

pyrit/executor/promptgen/fuzzer/fuzzer_expand_converter.py

        conversation_id = str(uuid.uuid4())

-        self.converter_target.set_system_prompt(
+        self.converter_target._set_target_system_prompt(


if we're calling this outside of target, it should be a public method

…orts, add seed model structural tests - Update PromptChatTarget to PromptTarget per PR microsoft#1532 deprecation - Move ScenarioStrategy import to top-level in test_foundry_contract.py - Add rationale for explicit inheritance checks in test_import_smoke.py - Expand seed model tests with structural validation (value, data_type, harm_categories, role, metadata, SeedGroup composition) Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

hannahwestra25 added 8 commits March 19, 2026 14:43

initial commit to remove prompt_chat_target

f80bc39

add validation check and keep chat wording

a31c550

remove all prompt chat target references

3034f91

remove set_system_prompt

04cdc00

remove unnecessary changes

da0cfe4

Merge branch 'main' into hawestra/remove_prompt_chat_target

e33cf6f

Merge branch 'main' of https://github.com/Azure/PyRIT into hawestra/r…

95a3ace

…emove_prompt_chat_target

Merge branch 'hawestra/remove_prompt_chat_target' of https://github.c…

b74280d

…om/hannahwestra25/PyRIT into hawestra/remove_prompt_chat_target

hannahwestra25 changed the title ~~[DRAFT] remove PromptChatTarget~~ MAINT remove PromptChatTarget Mar 23, 2026

hannahwestra25 added 2 commits March 23, 2026 15:49

pre-commit'

35aaa14

undo promptchattarget removal

f71950d

rlundeen2 reviewed Mar 24, 2026

View reviewed changes

hannahwestra25 mentioned this pull request Mar 24, 2026

FEAT: Add partner integration tests for azure-ai-evaluation red team … #1533

Open

rlundeen2 reviewed Mar 24, 2026

View reviewed changes

		@@ -56,7 +56,7 @@ async def convert_async(self, *, prompt: str, input_type: PromptDataType = "text

		conversation_id = str(uuid.uuid4())

Conversation

hannahwestra25 commented Mar 23, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Description

Tests and Documentation

Uh oh!

rlundeen2 Mar 24, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

rlundeen2 Mar 24, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

hannahwestra25 commented Mar 23, 2026 •

edited

Loading

rlundeen2 Mar 24, 2026 •

edited

Loading

rlundeen2 Mar 24, 2026 •

edited

Loading