Skip to content

DeepTeam is a framework to red team LLMs and AI agents.

finished linear jailbreaking

A-Vamshi7cd8a19fdf27
diff --git a/deepteam/attacks/multi_turn/linear_jailbreaking/linear_jailbreaking.py b/deepteam/attacks/multi_turn/linear_jailbreaking/linear_jailbreaking.py
index 807cfb64..014044de 100644
--- a/deepteam/attacks/multi_turn/linear_jailbreaking/linear_jailbreaking.py
+++ b/deepteam/attacks/multi_turn/linear_jailbreaking/linear_jailbreaking.py
@@ -210,7 +210,7 @@ class LinearJailbreaking(BaseAttack):
model_callback: CallbackType,
simulator_model: Optional[Union[DeepEvalBaseLLM, str]] = None,
turns: Optional[List[RTTurn]] = None,
- ) -> Dict[VulnerabilityType, List[RTTurn]]:
+ ) -> Dict[VulnerabilityType, List[List[RTTurn]]]:
from deepteam.red_teamer.utils import group_attacks_by_vulnerability_type
# Simulate and group attacks
simulated_attacks = group_attacks_by_vulnerability_type(
@@ -265,7 +265,9 @@ class LinearJailbreaking(BaseAttack):
vulnerability_type=vuln_type
)
- result[vuln_type] = enhanced_turns
+ attack.turn_history = enhanced_turns
+
+ result[vuln_type] = [attack.turn_history for attack in attacks]
return result
@@ -275,7 +277,7 @@ class LinearJailbreaking(BaseAttack):
model_callback: CallbackType,
simulator_model: Optional[Union[DeepEvalBaseLLM, str]] = None,
turns: Optional[List[RTTurn]] = None,
- ) -> Dict[VulnerabilityType, List[RTTurn]]:
+ ) -> Dict[VulnerabilityType, List[List[RTTurn]]]:
from deepteam.red_teamer.utils import group_attacks_by_vulnerability_type
# Simulate and group attacks asynchronously
@@ -285,7 +287,7 @@ class LinearJailbreaking(BaseAttack):
result = {}
for vuln_type, attacks in grouped_attacks.items():
- async def get_enhance_turns(attack):
+ async def enhance_attack(attack):
# Defensive copy of base turns
inner_turns = list(turns) if turns else []
@@ -321,7 +323,7 @@ class LinearJailbreaking(BaseAttack):
# Run async enhancement and store turn history
vulnerability_name = vulnerability.get_name()
- enhanced_turns = await self._a_get_turns(
+ attack.turn_history = await self._a_get_turns(
model_callback=model_callback,
turns=inner_turns,
simulator_model=simulator_model,
@@ -329,13 +331,13 @@ class LinearJailbreaking(BaseAttack):
vulnerability_type=vuln_type
)
- return enhanced_turns
+ return attack
# Run all attacks in this vulnerability group concurrently
- enhanced_turns = await asyncio.gather(
- *(get_enhance_turns(attack) for attack in attacks)
+ enhanced_attacks = await asyncio.gather(
+ *(enhance_attack(attack) for attack in attacks)
)
- result[vuln_type] = enhanced_turns
+ result[vuln_type] = [enhanced_attack.turn_history for enhanced_attack in enhanced_attacks]
return result