Skip to content

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

.

A-Vamshi84d0a8af9d3c
diff --git a/b.py b/b.py
index 89cb3bc7..435aa5c4 100644
--- a/b.py
+++ b/b.py
@@ -14,7 +14,7 @@ async def model_callback(attack, turn_history):
red_team(
- attacks=[LinearJailbreaking()],
+ attacks=[LinearJailbreaking(num_turns=3)],
vulnerabilities=[Bias(types=["race"], simulator_model="gpt-4o")],
model_callback=model_callback,
ignore_errors=False,
diff --git a/deepteam/attacks/multi_turn/linear_jailbreaking/linear_jailbreaking.py b/deepteam/attacks/multi_turn/linear_jailbreaking/linear_jailbreaking.py
index 4e397ce0..8b0e3214 100644
--- a/deepteam/attacks/multi_turn/linear_jailbreaking/linear_jailbreaking.py
+++ b/deepteam/attacks/multi_turn/linear_jailbreaking/linear_jailbreaking.py
@@ -238,6 +238,7 @@ class LinearJailbreaking(BaseAttack):
turns.append(RTTurn(role="assistant", content=target_response))
# Main simulation loop
+ # TODO: This loop adds the first input inside the turns again
for i in range(self.num_turns):
target_response = await model_callback(current_attack, turns)
turns.append(RTTurn(role="user", content=current_attack))