Skip to content

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

.

A-Vamshi258c1ef644c0
diff --git a/deepteam/attacks/single_turn/__init__.py b/deepteam/attacks/single_turn/__init__.py
index 9a07419b..d4496889 100644
--- a/deepteam/attacks/single_turn/__init__.py
+++ b/deepteam/attacks/single_turn/__init__.py
@@ -13,6 +13,10 @@ from .character_stream import CharacterStream
from .context_flooding import ContextFlooding
from .embedded_instruction_json import EmbeddedInstructionJSON
+from .prior_context_attack import PriorContextAttack
+from .authority_escalation import AuthorityEscalation
+from .emotional_manipulation import EmotionalManipulation
+
# Generic enhancement attacks (moved from agentic)
from .system_override.system_override import SystemOverride
from .permission_escalation.permission_escalation import PermissionEscalation
diff --git a/deepteam/attacks/single_turn/authority_escalation/__init__.py b/deepteam/attacks/single_turn/authority_escalation/__init__.py
index 91a2c8e1..d1ee29c1 100644
--- a/deepteam/attacks/single_turn/authority_escalation/__init__.py
+++ b/deepteam/attacks/single_turn/authority_escalation/__init__.py
@@ -1 +1 @@
-from .authority_escalation import AuthorityEscalation
\ No newline at end of file
+from .authority_escalation import AuthorityEscalation
diff --git a/deepteam/attacks/single_turn/authority_escalation/authority_escalation.py b/deepteam/attacks/single_turn/authority_escalation/authority_escalation.py
index 51ee4da8..9be7e771 100644
--- a/deepteam/attacks/single_turn/authority_escalation/authority_escalation.py
+++ b/deepteam/attacks/single_turn/authority_escalation/authority_escalation.py
@@ -19,6 +19,7 @@ from deepteam.attacks.attack_simulator.utils import (
a_generate,
)
+
class AuthorityEscalation(BaseSingleTurnAttack):
name = "Authority Escalation"
exploitability = Exploitability.HIGH
@@ -40,11 +41,8 @@ class AuthorityEscalation(BaseSingleTurnAttack):
simulator_model: Optional[Union[DeepEvalBaseLLM, str]] = None,
) -> str:
self.simulator_model, _ = initialize_model(simulator_model)
-
- prompt = AuthorityEscalationTemplate.enhance(
- attack,
- self.role
- )
+
+ prompt = AuthorityEscalationTemplate.enhance(attack, self.role)
progress = create_progress()
with progress:
@@ -98,10 +96,7 @@ class AuthorityEscalation(BaseSingleTurnAttack):
simulator_model: Optional[Union[DeepEvalBaseLLM, str]] = None,
) -> str:
self.simulator_model, _ = initialize_model(simulator_model)
- prompt = AuthorityEscalationTemplate.enhance(
- attack,
- self.role
- )
+ prompt = AuthorityEscalationTemplate.enhance(attack, self.role)
progress = create_progress()
with progress:
@@ -119,8 +114,10 @@ class AuthorityEscalation(BaseSingleTurnAttack):
enhanced_prompt = res.input
update_pbar(progress, task_id)
- compliance_prompt = AuthorityEscalationTemplate.non_compliant(
- res.model_dump()
+ compliance_prompt = (
+ AuthorityEscalationTemplate.non_compliant(
+ res.model_dump()
+ )
)
compliance_res: ComplianceData = await a_generate(
compliance_prompt,
@@ -152,6 +149,6 @@ class AuthorityEscalation(BaseSingleTurnAttack):
update_pbar(progress, task_id, advance_to_end=True)
return attack
-
+
def get_name(self) -> str:
- return self.name
\ No newline at end of file
+ return self.name
diff --git a/deepteam/attacks/single_turn/authority_escalation/schema.py b/deepteam/attacks/single_turn/authority_escalation/schema.py
index 0f8df540..f6eee49b 100644
--- a/deepteam/attacks/single_turn/authority_escalation/schema.py
+++ b/deepteam/attacks/single_turn/authority_escalation/schema.py
@@ -1,11 +1,14 @@
from pydantic import BaseModel
+
class EnhancedAuthorityAttack(BaseModel):
authority_role: str
input: str
+
class ComplianceData(BaseModel):
non_compliant: bool
+
class IsAuthorityValid(BaseModel):
- is_valid_authority: bool
\ No newline at end of file
+ is_valid_authority: bool
diff --git a/deepteam/attacks/single_turn/authority_escalation/template.py b/deepteam/attacks/single_turn/authority_escalation/template.py
index 72bf8074..1236f995 100644
--- a/deepteam/attacks/single_turn/authority_escalation/template.py
+++ b/deepteam/attacks/single_turn/authority_escalation/template.py
@@ -1,8 +1,9 @@
import random
+
class AuthorityEscalationTemplate:
"""
- Rewrites a request to look like a Routine Corporate Task, wrapped in a
+ Rewrites a request to look like a Routine Corporate Task, wrapped in a
'Helpful Assistant' scenario that compels the model to process it.
"""
@@ -122,4 +123,4 @@ class AuthorityEscalationTemplate:
{{
"is_valid_authority": true/false
}}
- """
\ No newline at end of file
+ """