Skip to content

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

add evaluation examples and evaluation guidelines support

A-Vamshi70aed5bb9c4f
diff --git a/deepteam/vulnerabilities/agent_identity_abuse/agent_identity_abuse.py b/deepteam/vulnerabilities/agent_identity_abuse/agent_identity_abuse.py
index 6a19a947..6b8492ba 100644
--- a/deepteam/vulnerabilities/agent_identity_abuse/agent_identity_abuse.py
+++ b/deepteam/vulnerabilities/agent_identity_abuse/agent_identity_abuse.py
@@ -11,6 +11,7 @@ from deepteam.vulnerabilities import BaseVulnerability
from deepteam.vulnerabilities.agent_identity_abuse import AgentIdentityAbuseType
from deepteam.vulnerabilities.utils import validate_vulnerability_types
from deepteam.metrics import AgentIdentityAbuseMetric, BaseRedTeamingMetric
+from deepteam.metrics.types import EvaluationExample
from deepteam.attacks.multi_turn.types import CallbackType
from deepteam.test_case import RTTestCase
from deepteam.attacks.attack_simulator.schema import SyntheticDataList
@@ -41,6 +42,8 @@ class AgentIdentityAbuse(BaseVulnerability):
type.value for type in AgentIdentityAbuseType
],
purpose: Optional[str] = None,
+ evaluation_examples: Optional[List[EvaluationExample]] = None,
+ evaluation_guidelines: Optional[List[str]] = None,
):
enum_types = validate_vulnerability_types(
self.get_name(), types=types, allowed_type=AgentIdentityAbuseType
@@ -50,6 +53,8 @@ class AgentIdentityAbuse(BaseVulnerability):
self.simulator_model = simulator_model
self.evaluation_model = evaluation_model
self.purpose = purpose
+ self.evaluation_examples = evaluation_examples
+ self.evaluation_guidelines = evaluation_guidelines
super().__init__(types=enum_types)
def assess(
@@ -289,6 +294,8 @@ class AgentIdentityAbuse(BaseVulnerability):
model=self.evaluation_model,
async_mode=self.async_mode,
verbose_mode=self.verbose_mode,
+ evaluation_examples=self.evaluation_examples,
+ evaluation_guidelines=self.evaluation_guidelines,
)
def is_vulnerable(self) -> bool:
diff --git a/deepteam/vulnerabilities/autonomous_agent_drift/autonomous_agent_drift.py b/deepteam/vulnerabilities/autonomous_agent_drift/autonomous_agent_drift.py
index 9c9e73c3..6e8e1447 100644
--- a/deepteam/vulnerabilities/autonomous_agent_drift/autonomous_agent_drift.py
+++ b/deepteam/vulnerabilities/autonomous_agent_drift/autonomous_agent_drift.py
@@ -13,6 +13,7 @@ from deepteam.vulnerabilities.autonomous_agent_drift import (
)
from deepteam.vulnerabilities.utils import validate_vulnerability_types
from deepteam.metrics import AutonomousAgentDriftMetric, BaseRedTeamingMetric
+from deepteam.metrics.types import EvaluationExample
from deepteam.attacks.multi_turn.types import CallbackType
from deepteam.test_case import RTTestCase
from deepteam.attacks.attack_simulator.schema import SyntheticDataList
@@ -46,6 +47,8 @@ class AutonomousAgentDrift(BaseVulnerability):
type.value for type in AutonomousAgentDriftType
],
purpose: Optional[str] = None,
+ evaluation_examples: Optional[List[EvaluationExample]] = None,
+ evaluation_guidelines: Optional[List[str]] = None,
):
enum_types = validate_vulnerability_types(
self.get_name(), types=types, allowed_type=AutonomousAgentDriftType
@@ -55,6 +58,8 @@ class AutonomousAgentDrift(BaseVulnerability):
self.simulator_model = simulator_model
self.evaluation_model = evaluation_model
self.purpose = purpose
+ self.evaluation_examples = evaluation_examples
+ self.evaluation_guidelines = evaluation_guidelines
super().__init__(types=enum_types)
def assess(
@@ -294,6 +299,8 @@ class AutonomousAgentDrift(BaseVulnerability):
model=self.evaluation_model,
async_mode=self.async_mode,
verbose_mode=self.verbose_mode,
+ evaluation_examples=self.evaluation_examples,
+ evaluation_guidelines=self.evaluation_guidelines,
)
def is_vulnerable(self) -> bool:
diff --git a/deepteam/vulnerabilities/bfla/bfla.py b/deepteam/vulnerabilities/bfla/bfla.py
index ff1793c2..8dd1a017 100644
--- a/deepteam/vulnerabilities/bfla/bfla.py
+++ b/deepteam/vulnerabilities/bfla/bfla.py
@@ -11,6 +11,7 @@ from deepteam.vulnerabilities import BaseVulnerability
from deepteam.vulnerabilities.bfla import BFLAType
from deepteam.vulnerabilities.utils import validate_vulnerability_types
from deepteam.metrics import BFLAMetric, BaseRedTeamingMetric
+from deepteam.metrics.types import EvaluationExample
from deepteam.attacks.multi_turn.types import CallbackType
from deepteam.test_case import RTTestCase
from deepteam.attacks.attack_simulator.schema import SyntheticDataList
@@ -39,6 +40,8 @@ class BFLA(BaseVulnerability):
evaluation_model: Optional[Union[str, DeepEvalBaseLLM]] = "gpt-4o",
types: Optional[List[BFLALiteral]] = [type.value for type in BFLAType],
purpose: Optional[str] = None,
+ evaluation_examples: Optional[List[EvaluationExample]] = None,
+ evaluation_guidelines: Optional[List[str]] = None,
):
enum_types = validate_vulnerability_types(
self.get_name(), types=types, allowed_type=BFLAType
@@ -48,6 +51,8 @@ class BFLA(BaseVulnerability):
self.simulator_model = simulator_model
self.evaluation_model = evaluation_model
self.purpose = purpose
+ self.evaluation_examples = evaluation_examples
+ self.evaluation_guidelines = evaluation_guidelines
super().__init__(types=enum_types)
def assess(
@@ -281,6 +286,8 @@ class BFLA(BaseVulnerability):
model=self.evaluation_model,
async_mode=self.async_mode,
verbose_mode=self.verbose_mode,
+ evaluation_examples=self.evaluation_examples,
+ evaluation_guidelines=self.evaluation_guidelines,
)
def is_vulnerable(self) -> bool:
diff --git a/deepteam/vulnerabilities/bias/bias.py b/deepteam/vulnerabilities/bias/bias.py
index 0ff4722c..478f74dd 100644
--- a/deepteam/vulnerabilities/bias/bias.py
+++ b/deepteam/vulnerabilities/bias/bias.py
@@ -11,6 +11,7 @@ from deepteam.attacks.base_attack import Exploitability
from deepteam.vulnerabilities.bias import BiasType
from deepteam.vulnerabilities.utils import validate_vulnerability_types
from deepteam.metrics import BiasMetric, BaseRedTeamingMetric
+from deepteam.metrics.types import EvaluationExample
from deepteam.attacks.multi_turn.types import CallbackType
from deepteam.test_case import RTTestCase
from deepteam.attacks.attack_simulator.schema import SyntheticDataList
@@ -37,6 +38,8 @@ class Bias(BaseVulnerability):
type.value for type in BiasType
],
purpose: Optional[str] = None,
+ evaluation_examples: Optional[List[EvaluationExample]] = None,
+ evaluation_guidelines: Optional[List[str]] = None,
):
enum_types = validate_vulnerability_types(
self.get_name(), types=types, allowed_type=BiasType
@@ -46,6 +49,8 @@ class Bias(BaseVulnerability):
self.simulator_model = simulator_model
self.evaluation_model = evaluation_model
self.purpose = purpose
+ self.evaluation_examples = evaluation_examples
+ self.evaluation_guidelines = evaluation_guidelines
super().__init__(types=enum_types)
def assess(
@@ -284,6 +289,8 @@ class Bias(BaseVulnerability):
model=self.evaluation_model,
async_mode=self.async_mode,
verbose_mode=self.verbose_mode,
+ evaluation_examples=self.evaluation_examples,
+ evaluation_guidelines=self.evaluation_guidelines,
)
def is_vulnerable(self) -> bool:
diff --git a/deepteam/vulnerabilities/bola/bola.py b/deepteam/vulnerabilities/bola/bola.py
index 156378a0..8f79493c 100644
--- a/deepteam/vulnerabilities/bola/bola.py
+++ b/deepteam/vulnerabilities/bola/bola.py
@@ -11,6 +11,7 @@ from deepteam.vulnerabilities import BaseVulnerability
from deepteam.vulnerabilities.bola import BOLAType
from deepteam.vulnerabilities.utils import validate_vulnerability_types
from deepteam.metrics import BOLAMetric, BaseRedTeamingMetric
+from deepteam.metrics.types import EvaluationExample
from deepteam.attacks.multi_turn.types import CallbackType
from deepteam.test_case import RTTestCase
from deepteam.attacks.attack_simulator.schema import SyntheticDataList
@@ -39,6 +40,8 @@ class BOLA(BaseVulnerability):
evaluation_model: Optional[Union[str, DeepEvalBaseLLM]] = "gpt-4o",
types: Optional[List[BOLALiteral]] = [type.value for type in BOLAType],
purpose: Optional[str] = None,
+ evaluation_examples: Optional[List[EvaluationExample]] = None,
+ evaluation_guidelines: Optional[List[str]] = None,
):
enum_types = validate_vulnerability_types(
self.get_name(), types=types, allowed_type=BOLAType
@@ -48,6 +51,8 @@ class BOLA(BaseVulnerability):
self.simulator_model = simulator_model
self.evaluation_model = evaluation_model
self.purpose = purpose
+ self.evaluation_examples = evaluation_examples
+ self.evaluation_guidelines = evaluation_guidelines
super().__init__(types=enum_types)
def assess(
@@ -284,6 +289,8 @@ class BOLA(BaseVulnerability):
model=self.evaluation_model,
async_mode=self.async_mode,
verbose_mode=self.verbose_mode,
+ evaluation_examples=self.evaluation_examples,
+ evaluation_guidelines=self.evaluation_guidelines,
)
def is_vulnerable(self) -> bool:
diff --git a/deepteam/vulnerabilities/child_protection/child_protection.py b/deepteam/vulnerabilities/child_protection/child_protection.py
index e671a476..64dae9ad 100644
--- a/deepteam/vulnerabilities/child_protection/child_protection.py
+++ b/deepteam/vulnerabilities/child_protection/child_protection.py
@@ -10,6 +10,7 @@ from deepteam.vulnerabilities import BaseVulnerability
from deepteam.vulnerabilities.child_protection import ChildProtectionType
from deepteam.vulnerabilities.utils import validate_vulnerability_types
from deepteam.metrics import ChildProtectionMetric, BaseRedTeamingMetric
+from deepteam.metrics.types import EvaluationExample
from deepteam.attacks.multi_turn.types import CallbackType
from deepteam.test_case import RTTestCase
from deepteam.attacks.attack_simulator.schema import SyntheticDataList
@@ -38,6 +39,8 @@ class ChildProtection(BaseVulnerability):
type.value for type in ChildProtectionType
],
purpose: Optional[str] = None,
+ evaluation_examples: Optional[List[EvaluationExample]] = None,
+ evaluation_guidelines: Optional[List[str]] = None,
):
enum_types = validate_vulnerability_types(
self.get_name(), types=types, allowed_type=ChildProtectionType
@@ -47,6 +50,8 @@ class ChildProtection(BaseVulnerability):
self.simulator_model = simulator_model
self.evaluation_model = evaluation_model
self.purpose = purpose
+ self.evaluation_examples = evaluation_examples
+ self.evaluation_guidelines = evaluation_guidelines
super().__init__(types=enum_types)
def assess(
@@ -285,6 +290,8 @@ class ChildProtection(BaseVulnerability):
model=self.evaluation_model,
async_mode=self.async_mode,
verbose_mode=self.verbose_mode,
+ evaluation_examples=self.evaluation_examples,
+ evaluation_guidelines=self.evaluation_guidelines,
)
def is_vulnerable(self) -> bool:
diff --git a/deepteam/vulnerabilities/competition/competition.py b/deepteam/vulnerabilities/competition/competition.py
index e6c14e93..b14c60af 100644
--- a/deepteam/vulnerabilities/competition/competition.py
+++ b/deepteam/vulnerabilities/competition/competition.py
@@ -12,6 +12,7 @@ from deepteam.vulnerabilities import BaseVulnerability
from deepteam.vulnerabilities.competition import CompetitionType
from deepteam.vulnerabilities.utils import validate_vulnerability_types
from deepteam.metrics import CompetitorsMetric, BaseRedTeamingMetric
+from deepteam.metrics.types import EvaluationExample
from deepteam.attacks.multi_turn.types import CallbackType
from deepteam.test_case import RTTestCase
from deepteam.attacks.attack_simulator.schema import SyntheticDataList
@@ -43,6 +44,8 @@ class Competition(BaseVulnerability):
type.value for type in CompetitionType
],
purpose: Optional[str] = None,
+ evaluation_examples: Optional[List[EvaluationExample]] = None,
+ evaluation_guidelines: Optional[List[str]] = None,
):
enum_types = validate_vulnerability_types(
self.get_name(), types=types, allowed_type=CompetitionType
@@ -52,6 +55,8 @@ class Competition(BaseVulnerability):
self.simulator_model = simulator_model
self.evaluation_model = evaluation_model
self.purpose = purpose
+ self.evaluation_examples = evaluation_examples
+ self.evaluation_guidelines = evaluation_guidelines
super().__init__(types=enum_types)
def assess(
@@ -288,6 +293,8 @@ class Competition(BaseVulnerability):
model=self.evaluation_model,
async_mode=self.async_mode,
verbose_mode=self.verbose_mode,
+ evaluation_examples=self.evaluation_examples,
+ evaluation_guidelines=self.evaluation_guidelines,
)
def is_vulnerable(self) -> bool:
diff --git a/deepteam/vulnerabilities/cross_context_retrieval/cross_context_retrieval.py b/deepteam/vulnerabilities/cross_context_retrieval/cross_context_retrieval.py
index 67d97867..3a308b43 100644
--- a/deepteam/vulnerabilities/cross_context_retrieval/cross_context_retrieval.py
+++ b/deepteam/vulnerabilities/cross_context_retrieval/cross_context_retrieval.py
@@ -12,6 +12,7 @@ from deepteam.vulnerabilities.cross_context_retrieval import (
)
from deepteam.vulnerabilities.utils import validate_vulnerability_types
from deepteam.metrics import CrossContextRetrievalMetric, BaseRedTeamingMetric
+from deepteam.metrics.types import EvaluationExample
from deepteam.attacks.multi_turn.types import CallbackType
from deepteam.test_case import RTTestCase
from deepteam.attacks.attack_simulator.schema import SyntheticDataList
@@ -38,6 +39,8 @@ class CrossContextRetrieval(BaseVulnerability):
type.value for type in CrossContextRetrievalType
],
purpose: Optional[str] = None,
+ evaluation_examples: Optional[List[EvaluationExample]] = None,
+ evaluation_guidelines: Optional[List[str]] = None,
):
enum_types = validate_vulnerability_types(
self.get_name(), types=types, allowed_type=CrossContextRetrievalType
@@ -47,6 +50,8 @@ class CrossContextRetrieval(BaseVulnerability):
self.simulator_model = simulator_model
self.evaluation_model = evaluation_model
self.purpose = purpose
+ self.evaluation_examples = evaluation_examples
+ self.evaluation_guidelines = evaluation_guidelines
super().__init__(types=enum_types)
def assess(
@@ -286,6 +291,8 @@ class CrossContextRetrieval(BaseVulnerability):
model=self.evaluation_model,
async_mode=self.async_mode,
verbose_mode=self.verbose_mode,
+ evaluation_examples=self.evaluation_examples,
+ evaluation_guidelines=self.evaluation_guidelines,
)
def is_vulnerable(self) -> bool:
diff --git a/deepteam/vulnerabilities/custom/custom.py b/deepteam/vulnerabilities/custom/custom.py
index 2fdb4e82..678a8d11 100644
--- a/deepteam/vulnerabilities/custom/custom.py
+++ b/deepteam/vulnerabilities/custom/custom.py
@@ -10,6 +10,7 @@ from deepteam.utils import validate_model_callback_signature
from deepteam.vulnerabilities import BaseVulnerability
from deepteam.metrics import BaseRedTeamingMetric, HarmMetric
+from deepteam.metrics.types import EvaluationExample
from deepteam.attacks.multi_turn.types import CallbackType
from deepteam.attacks.attack_simulator.schema import SyntheticDataList
from deepteam.risks import getRiskCategory
@@ -34,6 +35,8 @@ class CustomVulnerability(BaseVulnerability):
Union[str, DeepEvalBaseLLM]
] = "gpt-3.5-turbo-0125",
evaluation_model: Optional[Union[str, DeepEvalBaseLLM]] = "gpt-4o",
+ evaluation_examples: Optional[List[EvaluationExample]] = None,
+ evaluation_guidelines: Optional[List[str]] = None,
):
self.name = name
@@ -53,6 +56,8 @@ class CustomVulnerability(BaseVulnerability):
self.criteria = criteria.strip()
self.simulator_model = simulator_model
self.evaluation_model = evaluation_model
+ self.evaluation_examples = evaluation_examples
+ self.evaluation_guidelines = evaluation_guidelines
self.async_mode = async_mode
self.verbose_mode = verbose_mode
self.metric = None
@@ -303,6 +308,8 @@ class CustomVulnerability(BaseVulnerability):
model=self.evaluation_model,
async_mode=self.async_mode,
verbose_mode=self.verbose_mode,
+ evaluation_examples=self.evaluation_examples,
+ evaluation_guidelines=self.evaluation_guidelines,
)
return self.metric
diff --git a/deepteam/vulnerabilities/debug_access/debug_access.py b/deepteam/vulnerabilities/debug_access/debug_access.py
index 19fbafdb..e1d6d797 100644
--- a/deepteam/vulnerabilities/debug_access/debug_access.py
+++ b/deepteam/vulnerabilities/debug_access/debug_access.py
@@ -11,6 +11,7 @@ from deepteam.vulnerabilities import BaseVulnerability
from deepteam.vulnerabilities.debug_access import DebugAccessType
from deepteam.vulnerabilities.utils import validate_vulnerability_types
from deepteam.metrics import DebugAccessMetric, BaseRedTeamingMetric
+from deepteam.metrics.types import EvaluationExample
from deepteam.attacks.multi_turn.types import CallbackType
from deepteam.test_case import RTTestCase
from deepteam.attacks.attack_simulator.schema import SyntheticDataList
@@ -41,6 +42,8 @@ class DebugAccess(BaseVulnerability):
type.value for type in DebugAccessType
],
purpose: Optional[str] = None,
+ evaluation_examples: Optional[List[EvaluationExample]] = None,
+ evaluation_guidelines: Optional[List[str]] = None,
):
enum_types = validate_vulnerability_types(
self.get_name(), types=types, allowed_type=DebugAccessType
@@ -50,6 +53,8 @@ class DebugAccess(BaseVulnerability):
self.simulator_model = simulator_model
self.evaluation_model = evaluation_model
self.purpose = purpose
+ self.evaluation_examples = evaluation_examples
+ self.evaluation_guidelines = evaluation_guidelines
super().__init__(types=enum_types)
def assess(
@@ -283,6 +288,8 @@ class DebugAccess(BaseVulnerability):
model=self.evaluation_model,
async_mode=self.async_mode,
verbose_mode=self.verbose_mode,
+ evaluation_examples=self.evaluation_examples,
+ evaluation_guidelines=self.evaluation_guidelines,
)
def is_vulnerable(self) -> bool:
diff --git a/deepteam/vulnerabilities/ethics/ethics.py b/deepteam/vulnerabilities/ethics/ethics.py
index 67a31f8b..9d21df94 100644
--- a/deepteam/vulnerabilities/ethics/ethics.py
+++ b/deepteam/vulnerabilities/ethics/ethics.py
@@ -10,6 +10,7 @@ from deepteam.vulnerabilities import BaseVulnerability
from deepteam.vulnerabilities.ethics import EthicsType
from deepteam.vulnerabilities.utils import validate_vulnerability_types
from deepteam.metrics import EthicsMetric, BaseRedTeamingMetric
+from deepteam.metrics.types import EvaluationExample
from deepteam.attacks.multi_turn.types import CallbackType
from deepteam.test_case import RTTestCase
from deepteam.attacks.attack_simulator.schema import SyntheticDataList
@@ -38,6 +39,8 @@ class Ethics(BaseVulnerability):
type.value for type in EthicsType
],
purpose: Optional[str] = None,
+ evaluation_examples: Optional[List[EvaluationExample]] = None,
+ evaluation_guidelines: Optional[List[str]] = None,
):
enum_types = validate_vulnerability_types(
self.get_name(), types=types, allowed_type=EthicsType
@@ -47,6 +50,8 @@ class Ethics(BaseVulnerability):
self.simulator_model = simulator_model
self.evaluation_model = evaluation_model
self.purpose = purpose
+ self.evaluation_examples = evaluation_examples
+ self.evaluation_guidelines = evaluation_guidelines
super().__init__(types=enum_types)
def assess(
@@ -285,6 +290,8 @@ class Ethics(BaseVulnerability):
model=self.evaluation_model,
async_mode=self.async_mode,
verbose_mode=self.verbose_mode,
+ evaluation_examples=self.evaluation_examples,
+ evaluation_guidelines=self.evaluation_guidelines,
)
def is_vulnerable(self) -> bool:
diff --git a/deepteam/vulnerabilities/excessive_agency/excessive_agency.py b/deepteam/vulnerabilities/excessive_agency/excessive_agency.py
index 9c9aec33..d707413a 100644
--- a/deepteam/vulnerabilities/excessive_agency/excessive_agency.py
+++ b/deepteam/vulnerabilities/excessive_agency/excessive_agency.py
@@ -14,6 +14,7 @@ from deepteam.vulnerabilities.excessive_agency import (
from deepteam.test_case import RTTestCase
from deepteam.vulnerabilities.utils import validate_vulnerability_types
from deepteam.metrics import ExcessiveAgencyMetric, BaseRedTeamingMetric
+from deepteam.metrics.types import EvaluationExample
from deepteam.attacks.multi_turn.types import CallbackType
from deepteam.attacks.attack_simulator.schema import SyntheticDataList
from .template import ExcessiveAgencyTemplate
@@ -38,6 +39,8 @@ class ExcessiveAgency(BaseVulnerability):
type.value for type in ExcessiveAgencyType
],
purpose: Optional[str] = None,
+ evaluation_examples: Optional[List[EvaluationExample]] = None,
+ evaluation_guidelines: Optional[List[str]] = None,
):
enum_types = validate_vulnerability_types(
self.get_name(), types=types, allowed_type=ExcessiveAgencyType
@@ -47,6 +50,8 @@ class ExcessiveAgency(BaseVulnerability):
self.simulator_model = simulator_model
self.evaluation_model = evaluation_model
self.purpose = purpose
+ self.evaluation_examples = evaluation_examples
+ self.evaluation_guidelines = evaluation_guidelines
super().__init__(types=enum_types)
def assess(
@@ -282,6 +287,8 @@ class ExcessiveAgency(BaseVulnerability):
model=self.evaluation_model,
async_mode=self.async_mode,
verbose_mode=self.verbose_mode,
+ evaluation_examples=self.evaluation_examples,
+ evaluation_guidelines=self.evaluation_guidelines,
)
def is_vulnerable(self) -> bool:
diff --git a/deepteam/vulnerabilities/exploit_tool_agent/exploit_tool_agent.py b/deepteam/vulnerabilities/exploit_tool_agent/exploit_tool_agent.py
index 5b2ca0bc..a9c029c4 100644
--- a/deepteam/vulnerabilities/exploit_tool_agent/exploit_tool_agent.py
+++ b/deepteam/vulnerabilities/exploit_tool_agent/exploit_tool_agent.py
@@ -10,6 +10,7 @@ from deepteam.vulnerabilities import BaseVulnerability
from deepteam.vulnerabilities.exploit_tool_agent import ExploitToolAgentType
from deepteam.vulnerabilities.utils import validate_vulnerability_types
from deepteam.metrics import ExploitToolAgentMetric, BaseRedTeamingMetric
+from deepteam.metrics.types import EvaluationExample
from deepteam.attacks.multi_turn.types import CallbackType
from deepteam.test_case import RTTestCase
from deepteam.attacks.attack_simulator.schema import SyntheticDataList
@@ -38,6 +39,8 @@ class ExploitToolAgent(BaseVulnerability):
type.value for type in ExploitToolAgentType
],
purpose: Optional[str] = None,
+ evaluation_examples: Optional[List[EvaluationExample]] = None,
+ evaluation_guidelines: Optional[List[str]] = None,
):
enum_types = validate_vulnerability_types(
self.get_name(), types=types, allowed_type=ExploitToolAgentType
@@ -47,6 +50,8 @@ class ExploitToolAgent(BaseVulnerability):
self.simulator_model = simulator_model
self.evaluation_model = evaluation_model
self.purpose = purpose
+ self.evaluation_examples = evaluation_examples
+ self.evaluation_guidelines = evaluation_guidelines
super().__init__(types=enum_types)
def assess(
@@ -286,6 +291,8 @@ class ExploitToolAgent(BaseVulnerability):
model=self.evaluation_model,
async_mode=self.async_mode,
verbose_mode=self.verbose_mode,
+ evaluation_examples=self.evaluation_examples,
+ evaluation_guidelines=self.evaluation_guidelines,
)
def is_vulnerable(self) -> bool:
diff --git a/deepteam/vulnerabilities/external_system_abuse/external_system_abuse.py b/deepteam/vulnerabilities/external_system_abuse/external_system_abuse.py
index f3f6c170..94550e97 100644
--- a/deepteam/vulnerabilities/external_system_abuse/external_system_abuse.py
+++ b/deepteam/vulnerabilities/external_system_abuse/external_system_abuse.py
@@ -12,6 +12,7 @@ from deepteam.vulnerabilities.external_system_abuse import (
)
from deepteam.vulnerabilities.utils import validate_vulnerability_types
from deepteam.metrics import ExternalSystemAbuseMetric, BaseRedTeamingMetric
+from deepteam.metrics.types import EvaluationExample
from deepteam.attacks.multi_turn.types import CallbackType
from deepteam.test_case import RTTestCase
from deepteam.attacks.attack_simulator.schema import SyntheticDataList
@@ -40,6 +41,8 @@ class ExternalSystemAbuse(BaseVulnerability):
type.value for type in ExternalSystemAbuseType
],
purpose: Optional[str] = None,
+ evaluation_examples: Optional[List[EvaluationExample]] = None,
+ evaluation_guidelines: Optional[List[str]] = None,
):
enum_types = validate_vulnerability_types(
self.get_name(), types=types, allowed_type=ExternalSystemAbuseType
@@ -49,6 +52,8 @@ class ExternalSystemAbuse(BaseVulnerability):
self.simulator_model = simulator_model
self.evaluation_model = evaluation_model
self.purpose = purpose
+ self.evaluation_examples = evaluation_examples
+ self.evaluation_guidelines = evaluation_guidelines
super().__init__(types=enum_types)
def assess(
@@ -287,6 +292,8 @@ class ExternalSystemAbuse(BaseVulnerability):
model=self.evaluation_model,
async_mode=self.async_mode,
verbose_mode=self.verbose_mode,
+ evaluation_examples=self.evaluation_examples,
+ evaluation_guidelines=self.evaluation_guidelines,
)
def is_vulnerable(self) -> bool:
diff --git a/deepteam/vulnerabilities/fairness/fairness.py b/deepteam/vulnerabilities/fairness/fairness.py
index 87b36dcd..5d48725f 100644
--- a/deepteam/vulnerabilities/fairness/fairness.py
+++ b/deepteam/vulnerabilities/fairness/fairness.py
@@ -10,6 +10,7 @@ from deepteam.vulnerabilities import BaseVulnerability
from deepteam.vulnerabilities.fairness import FairnessType
from deepteam.vulnerabilities.utils import validate_vulnerability_types
from deepteam.metrics import FairnessMetric, BaseRedTeamingMetric
+from deepteam.metrics.types import EvaluationExample
from deepteam.attacks.multi_turn.types import CallbackType
from deepteam.test_case import RTTestCase
from deepteam.attacks.attack_simulator.schema import SyntheticDataList
@@ -38,6 +39,8 @@ class Fairness(BaseVulnerability):
type.value for type in FairnessType
],
purpose: Optional[str] = None,
+ evaluation_examples: Optional[List[EvaluationExample]] = None,
+ evaluation_guidelines: Optional[List[str]] = None,
):
enum_types = validate_vulnerability_types(
self.get_name(), types=types, allowed_type=FairnessType
@@ -47,6 +50,8 @@ class Fairness(BaseVulnerability):
self.simulator_model = simulator_model
self.evaluation_model = evaluation_model
self.purpose = purpose
+ self.evaluation_examples = evaluation_examples
+ self.evaluation_guidelines = evaluation_guidelines
super().__init__(types=enum_types)
def assess(
@@ -285,6 +290,8 @@ class Fairness(BaseVulnerability):
model=self.evaluation_model,
async_mode=self.async_mode,
verbose_mode=self.verbose_mode,
+ evaluation_examples=self.evaluation_examples,
+ evaluation_guidelines=self.evaluation_guidelines,
)
def is_vulnerable(self) -> bool:
diff --git a/deepteam/vulnerabilities/goal_theft/goal_theft.py b/deepteam/vulnerabilities/goal_theft/goal_theft.py
index afe41109..17c3ed98 100644
--- a/deepteam/vulnerabilities/goal_theft/goal_theft.py
+++ b/deepteam/vulnerabilities/goal_theft/goal_theft.py
@@ -1,3 +1,4 @@
+from deepteam.metrics.types import EvaluationExample
from typing import List, Literal, Optional, Union, Dict
import asyncio
@@ -41,6 +42,8 @@ class GoalTheft(BaseVulnerability):
type.value for type in GoalTheftType
],
purpose: Optional[str] = None,
+ evaluation_examples: Optional[List[EvaluationExample]] = None,
+ evaluation_guidelines: Optional[List[str]] = None,
):
enum_types = validate_vulnerability_types(
self.get_name(), types=types, allowed_type=GoalTheftType
@@ -50,6 +53,8 @@ class GoalTheft(BaseVulnerability):
self.simulator_model = simulator_model
self.evaluation_model = evaluation_model
self.purpose = purpose
+ self.evaluation_examples = evaluation_examples
+ self.evaluation_guidelines = evaluation_guidelines
super().__init__(types=enum_types)
def assess(
@@ -285,6 +290,8 @@ class GoalTheft(BaseVulnerability):
model=self.evaluation_model,
async_mode=self.async_mode,
verbose_mode=self.verbose_mode,
+ evaluation_examples=self.evaluation_examples,
+ evaluation_guidelines=self.evaluation_guidelines,
)
def is_vulnerable(self) -> bool:
diff --git a/deepteam/vulnerabilities/graphic_content/graphic_content.py b/deepteam/vulnerabilities/graphic_content/graphic_content.py
index 761c38f8..68f68e01 100644
--- a/deepteam/vulnerabilities/graphic_content/graphic_content.py
+++ b/deepteam/vulnerabilities/graphic_content/graphic_content.py
@@ -11,6 +11,7 @@ from deepteam.vulnerabilities import BaseVulnerability
from deepteam.vulnerabilities.graphic_content import GraphicContentType
from deepteam.vulnerabilities.utils import validate_vulnerability_types
from deepteam.metrics import GraphicMetric, BaseRedTeamingMetric
+from deepteam.metrics.types import EvaluationExample
from deepteam.attacks.multi_turn.types import CallbackType
from deepteam.test_case import RTTestCase
from deepteam.attacks.attack_simulator.schema import SyntheticDataList
@@ -39,6 +40,8 @@ class GraphicContent(BaseVulnerability):
type.value for type in GraphicContentType
],
purpose: Optional[str] = None,
+ evaluation_examples: Optional[List[EvaluationExample]] = None,
+ evaluation_guidelines: Optional[List[str]] = None,
):
enum_types = validate_vulnerability_types(
self.get_name(), types=types, allowed_type=GraphicContentType
@@ -48,6 +51,8 @@ class GraphicContent(BaseVulnerability):
self.simulator_model = simulator_model
self.evaluation_model = evaluation_model
self.purpose = purpose
+ self.evaluation_examples = evaluation_examples
+ self.evaluation_guidelines = evaluation_guidelines
super().__init__(types=enum_types)
def assess(
@@ -286,6 +291,8 @@ class GraphicContent(BaseVulnerability):
model=self.evaluation_model,
async_mode=self.async_mode,
verbose_mode=self.verbose_mode,
+ evaluation_examples=self.evaluation_examples,
+ evaluation_guidelines=self.evaluation_guidelines,
)
def is_vulnerable(self) -> bool:
diff --git a/deepteam/vulnerabilities/hallucination/hallucination.py b/deepteam/vulnerabilities/hallucination/hallucination.py
index 638344ea..ed9eae6a 100644
--- a/deepteam/vulnerabilities/hallucination/hallucination.py
+++ b/deepteam/vulnerabilities/hallucination/hallucination.py
@@ -11,6 +11,7 @@ from deepteam.attacks.base_attack import Exploitability
from deepteam.vulnerabilities.hallucination.types import HallucinationType
from deepteam.vulnerabilities.utils import validate_vulnerability_types
from deepteam.metrics import HallucinationMetric, BaseRedTeamingMetric
+from deepteam.metrics.types import EvaluationExample
from deepteam.attacks.multi_turn.types import CallbackType
from deepteam.test_case import RTTestCase
from deepteam.attacks.attack_simulator.schema import SyntheticDataList
@@ -42,6 +43,8 @@ class Hallucination(BaseVulnerability):
type.value for type in HallucinationType
],
purpose: Optional[str] = None,
+ evaluation_examples: Optional[List[EvaluationExample]] = None,
+ evaluation_guidelines: Optional[List[str]] = None,
):
enum_types = validate_vulnerability_types(
self.get_name(), types=types, allowed_type=HallucinationType
@@ -51,6 +54,8 @@ class Hallucination(BaseVulnerability):
self.simulator_model = simulator_model
self.evaluation_model = evaluation_model
self.purpose = purpose
+ self.evaluation_examples = evaluation_examples
+ self.evaluation_guidelines = evaluation_guidelines
super().__init__(types=enum_types)
def assess(
@@ -280,6 +285,8 @@ class Hallucination(BaseVulnerability):
model=self.evaluation_model,
async_mode=self.async_mode,
verbose_mode=self.verbose_mode,
+ evaluation_examples=self.evaluation_examples,
+ evaluation_guidelines=self.evaluation_guidelines,
)
def is_vulnerable(self) -> bool:
diff --git a/deepteam/vulnerabilities/illegal_activity/illegal_activity.py b/deepteam/vulnerabilities/illegal_activity/illegal_activity.py
index d1e2c583..0c1bd244 100644
--- a/deepteam/vulnerabilities/illegal_activity/illegal_activity.py
+++ b/deepteam/vulnerabilities/illegal_activity/illegal_activity.py
@@ -11,6 +11,7 @@ from deepteam.vulnerabilities import BaseVulnerability
from deepteam.vulnerabilities.illegal_activity import IllegalActivityType
from deepteam.vulnerabilities.utils import validate_vulnerability_types
from deepteam.metrics import IllegalMetric, BaseRedTeamingMetric
+from deepteam.metrics.types import EvaluationExample
from deepteam.attacks.multi_turn.types import CallbackType
from deepteam.test_case import RTTestCase
from deepteam.attacks.attack_simulator.schema import SyntheticDataList
@@ -45,6 +46,8 @@ class IllegalActivity(BaseVulnerability):
type.value for type in IllegalActivityType
],
purpose: Optional[str] = None,
+ evaluation_examples: Optional[List[EvaluationExample]] = None,
+ evaluation_guidelines: Optional[List[str]] = None,
):
enum_types = validate_vulnerability_types(
self.get_name(), types=types, allowed_type=IllegalActivityType
@@ -54,6 +57,8 @@ class IllegalActivity(BaseVulnerability):
self.simulator_model = simulator_model
self.evaluation_model = evaluation_model
self.purpose = purpose
+ self.evaluation_examples = evaluation_examples
+ self.evaluation_guidelines = evaluation_guidelines
super().__init__(types=enum_types)
def assess(
@@ -293,6 +298,8 @@ class IllegalActivity(BaseVulnerability):
model=self.evaluation_model,
async_mode=self.async_mode,
verbose_mode=self.verbose_mode,
+ evaluation_examples=self.evaluation_examples,
+ evaluation_guidelines=self.evaluation_guidelines,
)
def is_vulnerable(self) -> bool:
diff --git a/deepteam/vulnerabilities/indirect_instruction/indirect_instruction.py b/deepteam/vulnerabilities/indirect_instruction/indirect_instruction.py
index 46523fa1..f81f2c26 100644
--- a/deepteam/vulnerabilities/indirect_instruction/indirect_instruction.py
+++ b/deepteam/vulnerabilities/indirect_instruction/indirect_instruction.py
@@ -13,6 +13,7 @@ from deepteam.vulnerabilities.indirect_instruction import (
)
from deepteam.vulnerabilities.utils import validate_vulnerability_types
from deepteam.metrics import IndirectInstructionMetric, BaseRedTeamingMetric
+from deepteam.metrics.types import EvaluationExample
from deepteam.attacks.multi_turn.types import CallbackType
from deepteam.test_case import RTTestCase
from deepteam.attacks.attack_simulator.schema import SyntheticDataList
@@ -44,6 +45,8 @@ class IndirectInstruction(BaseVulnerability):
type.value for type in IndirectInstructionType
],
purpose: Optional[str] = None,
+ evaluation_examples: Optional[List[EvaluationExample]] = None,
+ evaluation_guidelines: Optional[List[str]] = None,
):
enum_types = validate_vulnerability_types(
self.get_name(), types=types, allowed_type=IndirectInstructionType
@@ -53,6 +56,8 @@ class IndirectInstruction(BaseVulnerability):
self.simulator_model = simulator_model
self.evaluation_model = evaluation_model
self.purpose = purpose
+ self.evaluation_examples = evaluation_examples
+ self.evaluation_guidelines = evaluation_guidelines
super().__init__(types=enum_types)
def assess(
@@ -292,6 +297,8 @@ class IndirectInstruction(BaseVulnerability):
model=self.evaluation_model,
async_mode=self.async_mode,
verbose_mode=self.verbose_mode,
+ evaluation_examples=self.evaluation_examples,
+ evaluation_guidelines=self.evaluation_guidelines,
)
def is_vulnerable(self) -> bool:
diff --git a/deepteam/vulnerabilities/insecure_inter_agent_communication/insecure_inter_agent_communication.py b/deepteam/vulnerabilities/insecure_inter_agent_communication/insecure_inter_agent_communication.py
index f1b3b2bb..54933a7f 100644
--- a/deepteam/vulnerabilities/insecure_inter_agent_communication/insecure_inter_agent_communication.py
+++ b/deepteam/vulnerabilities/insecure_inter_agent_communication/insecure_inter_agent_communication.py
@@ -16,6 +16,7 @@ from deepteam.metrics import (
InsecureInterAgentCommunicationMetric,
BaseRedTeamingMetric,
)
+from deepteam.metrics.types import EvaluationExample
from deepteam.attacks.multi_turn.types import CallbackType
from deepteam.test_case import RTTestCase
from deepteam.attacks.attack_simulator.schema import SyntheticDataList
@@ -44,6 +45,8 @@ class InsecureInterAgentCommunication(BaseVulnerability):
type.value for type in InsecureInterAgentCommunicationType
],
purpose: Optional[str] = None,
+ evaluation_examples: Optional[List[EvaluationExample]] = None,
+ evaluation_guidelines: Optional[List[str]] = None,
):
enum_types = validate_vulnerability_types(
self.get_name(),
@@ -55,6 +58,8 @@ class InsecureInterAgentCommunication(BaseVulnerability):
self.simulator_model = simulator_model
self.evaluation_model = evaluation_model
self.purpose = purpose
+ self.evaluation_examples = evaluation_examples
+ self.evaluation_guidelines = evaluation_guidelines
super().__init__(types=enum_types)
def assess(
@@ -304,6 +309,8 @@ class InsecureInterAgentCommunication(BaseVulnerability):
model=self.evaluation_model,
async_mode=self.async_mode,
verbose_mode=self.verbose_mode,
+ evaluation_examples=self.evaluation_examples,
+ evaluation_guidelines=self.evaluation_guidelines,
)
def is_vulnerable(self) -> bool:
diff --git a/deepteam/vulnerabilities/intellectual_property/intellectual_property.py b/deepteam/vulnerabilities/intellectual_property/intellectual_property.py
index 151c7cf1..83e50b40 100644
--- a/deepteam/vulnerabilities/intellectual_property/intellectual_property.py
+++ b/deepteam/vulnerabilities/intellectual_property/intellectual_property.py
@@ -13,6 +13,7 @@ from deepteam.vulnerabilities.intellectual_property import (
)
from deepteam.vulnerabilities.utils import validate_vulnerability_types
from deepteam.metrics import IntellectualPropertyMetric, BaseRedTeamingMetric
+from deepteam.metrics.types import EvaluationExample
from deepteam.attacks.multi_turn.types import CallbackType
from deepteam.test_case import RTTestCase
from deepteam.attacks.attack_simulator.schema import SyntheticDataList
@@ -44,6 +45,8 @@ class IntellectualProperty(BaseVulnerability):
type.value for type in IntellectualPropertyType
],
purpose: Optional[str] = None,
+ evaluation_examples: Optional[List[EvaluationExample]] = None,
+ evaluation_guidelines: Optional[List[str]] = None,
):
enum_types = validate_vulnerability_types(
self.get_name(), types=types, allowed_type=IntellectualPropertyType
@@ -53,6 +56,8 @@ class IntellectualProperty(BaseVulnerability):
self.simulator_model = simulator_model
self.evaluation_model = evaluation_model
self.purpose = purpose
+ self.evaluation_examples = evaluation_examples
+ self.evaluation_guidelines = evaluation_guidelines
super().__init__(types=enum_types)
def assess(
@@ -290,6 +295,8 @@ class IntellectualProperty(BaseVulnerability):
model=self.evaluation_model,
async_mode=self.async_mode,
verbose_mode=self.verbose_mode,
+ evaluation_examples=self.evaluation_examples,
+ evaluation_guidelines=self.evaluation_guidelines,
)
def is_vulnerable(self) -> bool:
diff --git a/deepteam/vulnerabilities/misinformation/misinformation.py b/deepteam/vulnerabilities/misinformation/misinformation.py
index 453f4f44..2c96873a 100644
--- a/deepteam/vulnerabilities/misinformation/misinformation.py
+++ b/deepteam/vulnerabilities/misinformation/misinformation.py
@@ -11,6 +11,7 @@ from deepteam.vulnerabilities import BaseVulnerability
from deepteam.vulnerabilities.misinformation import MisinformationType
from deepteam.vulnerabilities.utils import validate_vulnerability_types
from deepteam.metrics import MisinformationMetric, BaseRedTeamingMetric
+from deepteam.metrics.types import EvaluationExample
from deepteam.attacks.multi_turn.types import CallbackType
from deepteam.test_case import RTTestCase
from deepteam.attacks.attack_simulator.schema import SyntheticDataList
@@ -39,6 +40,8 @@ class Misinformation(BaseVulnerability):
type.value for type in MisinformationType
],
purpose: Optional[str] = None,
+ evaluation_examples: Optional[List[EvaluationExample]] = None,
+ evaluation_guidelines: Optional[List[str]] = None,
):
enum_types = validate_vulnerability_types(
self.get_name(), types=types, allowed_type=MisinformationType
@@ -48,6 +51,8 @@ class Misinformation(BaseVulnerability):
self.simulator_model = simulator_model
self.evaluation_model = evaluation_model
self.purpose = purpose
+ self.evaluation_examples = evaluation_examples
+ self.evaluation_guidelines = evaluation_guidelines
super().__init__(types=enum_types)
def assess(
@@ -285,6 +290,8 @@ class Misinformation(BaseVulnerability):
model=self.evaluation_model,
async_mode=self.async_mode,
verbose_mode=self.verbose_mode,
+ evaluation_examples=self.evaluation_examples,
+ evaluation_guidelines=self.evaluation_guidelines,
)
def is_vulnerable(self) -> bool:
diff --git a/deepteam/vulnerabilities/personal_safety/personal_safety.py b/deepteam/vulnerabilities/personal_safety/personal_safety.py
index 59b00b1b..1dd25a4b 100644
--- a/deepteam/vulnerabilities/personal_safety/personal_safety.py
+++ b/deepteam/vulnerabilities/personal_safety/personal_safety.py
@@ -11,6 +11,7 @@ from deepteam.vulnerabilities import BaseVulnerability
from deepteam.vulnerabilities.personal_safety import PersonalSafetyType
from deepteam.vulnerabilities.utils import validate_vulnerability_types
from deepteam.metrics import SafetyMetric, BaseRedTeamingMetric
+from deepteam.metrics.types import EvaluationExample
from deepteam.attacks.multi_turn.types import CallbackType
from deepteam.test_case import RTTestCase
from deepteam.attacks.attack_simulator.schema import SyntheticDataList
@@ -43,6 +44,8 @@ class PersonalSafety(BaseVulnerability):
type.value for type in PersonalSafetyType
],
purpose: Optional[str] = None,
+ evaluation_examples: Optional[List[EvaluationExample]] = None,
+ evaluation_guidelines: Optional[List[str]] = None,
):
enum_types = validate_vulnerability_types(
self.get_name(), types=types, allowed_type=PersonalSafetyType
@@ -52,6 +55,8 @@ class PersonalSafety(BaseVulnerability):
self.simulator_model = simulator_model
self.evaluation_model = evaluation_model
self.purpose = purpose
+ self.evaluation_examples = evaluation_examples
+ self.evaluation_guidelines = evaluation_guidelines
super().__init__(types=enum_types)
def assess(
@@ -285,6 +290,8 @@ class PersonalSafety(BaseVulnerability):
model=self.evaluation_model,
async_mode=self.async_mode,
verbose_mode=self.verbose_mode,
+ evaluation_examples=self.evaluation_examples,
+ evaluation_guidelines=self.evaluation_guidelines,
)
def is_vulnerable(self) -> bool:
diff --git a/deepteam/vulnerabilities/pii_leakage/pii_leakage.py b/deepteam/vulnerabilities/pii_leakage/pii_leakage.py
index 01667983..1413b4ec 100644
--- a/deepteam/vulnerabilities/pii_leakage/pii_leakage.py
+++ b/deepteam/vulnerabilities/pii_leakage/pii_leakage.py
@@ -11,6 +11,7 @@ from deepteam.vulnerabilities import BaseVulnerability
from deepteam.vulnerabilities.pii_leakage import PIILeakageType
from deepteam.vulnerabilities.utils import validate_vulnerability_types
from deepteam.metrics import PIIMetric, BaseRedTeamingMetric
+from deepteam.metrics.types import EvaluationExample
from deepteam.attacks.multi_turn.types import CallbackType
from deepteam.test_case import RTTestCase
from deepteam.attacks.attack_simulator.schema import SyntheticDataList
@@ -42,6 +43,8 @@ class PIILeakage(BaseVulnerability):
type.value for type in PIILeakageType
],
purpose: Optional[str] = None,
+ evaluation_examples: Optional[List[EvaluationExample]] = None,
+ evaluation_guidelines: Optional[List[str]] = None,
):
enum_types = validate_vulnerability_types(
self.get_name(), types=types, allowed_type=PIILeakageType
@@ -51,6 +54,8 @@ class PIILeakage(BaseVulnerability):
self.simulator_model = simulator_model
self.evaluation_model = evaluation_model
self.purpose = purpose
+ self.evaluation_examples = evaluation_examples
+ self.evaluation_guidelines = evaluation_guidelines
super().__init__(types=enum_types)
def assess(
@@ -291,6 +296,8 @@ class PIILeakage(BaseVulnerability):
model=self.evaluation_model,
async_mode=self.async_mode,
verbose_mode=self.verbose_mode,
+ evaluation_examples=self.evaluation_examples,
+ evaluation_guidelines=self.evaluation_guidelines,
)
def is_vulnerable(self) -> bool:
diff --git a/deepteam/vulnerabilities/prompt_leakage/prompt_leakage.py b/deepteam/vulnerabilities/prompt_leakage/prompt_leakage.py
index c0848e6b..6bf4482d 100644
--- a/deepteam/vulnerabilities/prompt_leakage/prompt_leakage.py
+++ b/deepteam/vulnerabilities/prompt_leakage/prompt_leakage.py
@@ -11,6 +11,7 @@ from deepteam.vulnerabilities import BaseVulnerability
from deepteam.vulnerabilities.prompt_leakage import PromptLeakageType
from deepteam.vulnerabilities.utils import validate_vulnerability_types
from deepteam.metrics import PromptExtractionMetric, BaseRedTeamingMetric
+from deepteam.metrics.types import EvaluationExample
from deepteam.attacks.multi_turn.types import CallbackType
from deepteam.test_case import RTTestCase
from deepteam.attacks.attack_simulator.schema import SyntheticDataList
@@ -42,6 +43,8 @@ class PromptLeakage(BaseVulnerability):
type.value for type in PromptLeakageType
],
purpose: Optional[str] = None,
+ evaluation_examples: Optional[List[EvaluationExample]] = None,
+ evaluation_guidelines: Optional[List[str]] = None,
):
enum_types = validate_vulnerability_types(
self.get_name(), types=types, allowed_type=PromptLeakageType
@@ -51,6 +54,8 @@ class PromptLeakage(BaseVulnerability):
self.simulator_model = simulator_model
self.evaluation_model = evaluation_model
self.purpose = purpose
+ self.evaluation_examples = evaluation_examples
+ self.evaluation_guidelines = evaluation_guidelines
super().__init__(types=enum_types)
def assess(
@@ -288,6 +293,8 @@ class PromptLeakage(BaseVulnerability):
model=self.evaluation_model,
async_mode=self.async_mode,
verbose_mode=self.verbose_mode,
+ evaluation_examples=self.evaluation_examples,
+ evaluation_guidelines=self.evaluation_guidelines,
)
def is_vulnerable(self) -> bool:
diff --git a/deepteam/vulnerabilities/rbac/rbac.py b/deepteam/vulnerabilities/rbac/rbac.py
index ca2ea6ac..adc4b669 100644
--- a/deepteam/vulnerabilities/rbac/rbac.py
+++ b/deepteam/vulnerabilities/rbac/rbac.py
@@ -11,6 +11,7 @@ from deepteam.vulnerabilities import BaseVulnerability
from deepteam.vulnerabilities.rbac import RBACType
from deepteam.vulnerabilities.utils import validate_vulnerability_types
from deepteam.metrics import RBACMetric, BaseRedTeamingMetric
+from deepteam.metrics.types import EvaluationExample
from deepteam.attacks.multi_turn.types import CallbackType
from deepteam.test_case import RTTestCase
from deepteam.attacks.attack_simulator.schema import SyntheticDataList
@@ -39,6 +40,8 @@ class RBAC(BaseVulnerability):
evaluation_model: Optional[Union[str, DeepEvalBaseLLM]] = "gpt-4o",
types: Optional[List[RBACLiteral]] = [type.value for type in RBACType],
purpose: Optional[str] = None,
+ evaluation_examples: Optional[List[EvaluationExample]] = None,
+ evaluation_guidelines: Optional[List[str]] = None,
):
enum_types = validate_vulnerability_types(
self.get_name(), types=types, allowed_type=RBACType
@@ -48,6 +51,8 @@ class RBAC(BaseVulnerability):
self.simulator_model = simulator_model
self.evaluation_model = evaluation_model
self.purpose = purpose
+ self.evaluation_examples = evaluation_examples
+ self.evaluation_guidelines = evaluation_guidelines
super().__init__(types=enum_types)
def assess(
@@ -285,6 +290,8 @@ class RBAC(BaseVulnerability):
model=self.evaluation_model,
async_mode=self.async_mode,
verbose_mode=self.verbose_mode,
+ evaluation_examples=self.evaluation_examples,
+ evaluation_guidelines=self.evaluation_guidelines,
)
def is_vulnerable(self) -> bool:
diff --git a/deepteam/vulnerabilities/recursive_hijacking/recursive_hijacking.py b/deepteam/vulnerabilities/recursive_hijacking/recursive_hijacking.py
index 8fa45042..bbb7f0d7 100644
--- a/deepteam/vulnerabilities/recursive_hijacking/recursive_hijacking.py
+++ b/deepteam/vulnerabilities/recursive_hijacking/recursive_hijacking.py
@@ -1,3 +1,4 @@
+from deepteam.metrics.types import EvaluationExample
from typing import List, Literal, Optional, Union, Dict
import asyncio
@@ -45,6 +46,8 @@ class RecursiveHijacking(BaseVulnerability):
type.value for type in RecursiveHijackingType
],
purpose: Optional[str] = None,
+ evaluation_examples: Optional[List[EvaluationExample]] = None,
+ evaluation_guidelines: Optional[List[str]] = None,
):
enum_types = validate_vulnerability_types(
self.get_name(), types=types, allowed_type=RecursiveHijackingType
@@ -54,6 +57,8 @@ class RecursiveHijacking(BaseVulnerability):
self.simulator_model = simulator_model
self.evaluation_model = evaluation_model
self.purpose = purpose
+ self.evaluation_examples = evaluation_examples
+ self.evaluation_guidelines = evaluation_guidelines
super().__init__(types=enum_types)
def assess(
@@ -289,6 +294,8 @@ class RecursiveHijacking(BaseVulnerability):
model=self.evaluation_model,
async_mode=self.async_mode,
verbose_mode=self.verbose_mode,
+ evaluation_examples=self.evaluation_examples,
+ evaluation_guidelines=self.evaluation_guidelines,
)
def is_vulnerable(self) -> bool:
diff --git a/deepteam/vulnerabilities/robustness/robustness.py b/deepteam/vulnerabilities/robustness/robustness.py
index bfe27891..7320265d 100644
--- a/deepteam/vulnerabilities/robustness/robustness.py
+++ b/deepteam/vulnerabilities/robustness/robustness.py
@@ -15,6 +15,7 @@ from deepteam.metrics import (
OverrelianceMetric,
BaseRedTeamingMetric,
)
+from deepteam.metrics.types import EvaluationExample
from deepteam.test_case import RTTestCase
from deepteam.attacks.multi_turn.types import CallbackType
from deepteam.attacks.attack_simulator.schema import SyntheticDataList
@@ -40,6 +41,8 @@ class Robustness(BaseVulnerability):
type.value for type in RobustnessType
],
purpose: Optional[str] = None,
+ evaluation_examples: Optional[List[EvaluationExample]] = None,
+ evaluation_guidelines: Optional[List[str]] = None,
):
enum_types = validate_vulnerability_types(
self.get_name(), types=types, allowed_type=RobustnessType
@@ -49,6 +52,8 @@ class Robustness(BaseVulnerability):
self.simulator_model = simulator_model
self.evaluation_model = evaluation_model
self.purpose = purpose
+ self.evaluation_examples = evaluation_examples
+ self.evaluation_guidelines = evaluation_guidelines
super().__init__(types=enum_types)
def assess(
@@ -286,6 +291,8 @@ class Robustness(BaseVulnerability):
model=self.evaluation_model,
async_mode=self.async_mode,
verbose_mode=self.verbose_mode,
+ evaluation_examples=self.evaluation_examples,
+ evaluation_guidelines=self.evaluation_guidelines,
)
if type == RobustnessType.INPUT_OVERRELIANCE:
return OverrelianceMetric(
@@ -293,6 +300,8 @@ class Robustness(BaseVulnerability):
model=self.evaluation_model,
async_mode=self.async_mode,
verbose_mode=self.verbose_mode,
+ evaluation_examples=self.evaluation_examples,
+ evaluation_guidelines=self.evaluation_guidelines,
)
raise ValueError(
"Invalid type passed in the 'get_metric' function. Please pass an enum from 'RobustnessType'"
diff --git a/deepteam/vulnerabilities/shell_injection/shell_injection.py b/deepteam/vulnerabilities/shell_injection/shell_injection.py
index 5680de76..b03c6618 100644
--- a/deepteam/vulnerabilities/shell_injection/shell_injection.py
+++ b/deepteam/vulnerabilities/shell_injection/shell_injection.py
@@ -11,6 +11,7 @@ from deepteam.vulnerabilities import BaseVulnerability
from deepteam.vulnerabilities.shell_injection import ShellInjectionType
from deepteam.vulnerabilities.utils import validate_vulnerability_types
from deepteam.metrics import ShellInjectionMetric, BaseRedTeamingMetric
+from deepteam.metrics.types import EvaluationExample
from deepteam.attacks.multi_turn.types import CallbackType
from deepteam.test_case import RTTestCase
from deepteam.attacks.attack_simulator.schema import SyntheticDataList
@@ -41,6 +42,8 @@ class ShellInjection(BaseVulnerability):
type.value for type in ShellInjectionType
],
purpose: Optional[str] = None,
+ evaluation_examples: Optional[List[EvaluationExample]] = None,
+ evaluation_guidelines: Optional[List[str]] = None,
):
enum_types = validate_vulnerability_types(
self.get_name(), types=types, allowed_type=ShellInjectionType
@@ -50,6 +53,8 @@ class ShellInjection(BaseVulnerability):
self.simulator_model = simulator_model
self.evaluation_model = evaluation_model
self.purpose = purpose
+ self.evaluation_examples = evaluation_examples
+ self.evaluation_guidelines = evaluation_guidelines
super().__init__(types=enum_types)
def assess(
@@ -280,6 +285,8 @@ class ShellInjection(BaseVulnerability):
model=self.evaluation_model,
async_mode=self.async_mode,
verbose_mode=self.verbose_mode,
+ evaluation_examples=self.evaluation_examples,
+ evaluation_guidelines=self.evaluation_guidelines,
)
def is_vulnerable(self) -> bool:
diff --git a/deepteam/vulnerabilities/sql_injection/sql_injection.py b/deepteam/vulnerabilities/sql_injection/sql_injection.py
index 55d8a7cc..41191b99 100644
--- a/deepteam/vulnerabilities/sql_injection/sql_injection.py
+++ b/deepteam/vulnerabilities/sql_injection/sql_injection.py
@@ -11,6 +11,7 @@ from deepteam.vulnerabilities import BaseVulnerability
from deepteam.vulnerabilities.sql_injection import SQLInjectionType
from deepteam.vulnerabilities.utils import validate_vulnerability_types
from deepteam.metrics import SQLInjectionMetric, BaseRedTeamingMetric
+from deepteam.metrics.types import EvaluationExample
from deepteam.attacks.multi_turn.types import CallbackType
from deepteam.test_case import RTTestCase
from deepteam.attacks.attack_simulator.schema import SyntheticDataList
@@ -41,6 +42,8 @@ class SQLInjection(BaseVulnerability):
type.value for type in SQLInjectionType
],
purpose: Optional[str] = None,
+ evaluation_examples: Optional[List[EvaluationExample]] = None,
+ evaluation_guidelines: Optional[List[str]] = None,
):
enum_types = validate_vulnerability_types(
self.get_name(), types=types, allowed_type=SQLInjectionType
@@ -50,6 +53,8 @@ class SQLInjection(BaseVulnerability):
self.simulator_model = simulator_model
self.evaluation_model = evaluation_model
self.purpose = purpose
+ self.evaluation_examples = evaluation_examples
+ self.evaluation_guidelines = evaluation_guidelines
super().__init__(types=enum_types)
def assess(
@@ -280,6 +285,8 @@ class SQLInjection(BaseVulnerability):
model=self.evaluation_model,
async_mode=self.async_mode,
verbose_mode=self.verbose_mode,
+ evaluation_examples=self.evaluation_examples,
+ evaluation_guidelines=self.evaluation_guidelines,
)
def is_vulnerable(self) -> bool:
diff --git a/deepteam/vulnerabilities/ssrf/ssrf.py b/deepteam/vulnerabilities/ssrf/ssrf.py
index 3077e38d..139f275a 100644
--- a/deepteam/vulnerabilities/ssrf/ssrf.py
+++ b/deepteam/vulnerabilities/ssrf/ssrf.py
@@ -11,6 +11,7 @@ from deepteam.vulnerabilities import BaseVulnerability
from deepteam.vulnerabilities.ssrf import SSRFType
from deepteam.vulnerabilities.utils import validate_vulnerability_types
from deepteam.metrics import SSRFMetric, BaseRedTeamingMetric
+from deepteam.metrics.types import EvaluationExample
from deepteam.attacks.multi_turn.types import CallbackType
from deepteam.test_case import RTTestCase
from deepteam.attacks.attack_simulator.schema import SyntheticDataList
@@ -39,6 +40,8 @@ class SSRF(BaseVulnerability):
evaluation_model: Optional[Union[str, DeepEvalBaseLLM]] = "gpt-4o",
types: Optional[List[SSRFLiteral]] = [type.value for type in SSRFType],
purpose: Optional[str] = None,
+ evaluation_examples: Optional[List[EvaluationExample]] = None,
+ evaluation_guidelines: Optional[List[str]] = None,
):
enum_types = validate_vulnerability_types(
self.get_name(), types=types, allowed_type=SSRFType
@@ -48,6 +51,8 @@ class SSRF(BaseVulnerability):
self.simulator_model = simulator_model
self.evaluation_model = evaluation_model
self.purpose = purpose
+ self.evaluation_examples = evaluation_examples
+ self.evaluation_guidelines = evaluation_guidelines
super().__init__(types=enum_types)
def assess(
@@ -279,6 +284,8 @@ class SSRF(BaseVulnerability):
model=self.evaluation_model,
async_mode=self.async_mode,
verbose_mode=self.verbose_mode,
+ evaluation_examples=self.evaluation_examples,
+ evaluation_guidelines=self.evaluation_guidelines,
)
def is_vulnerable(self) -> bool:
diff --git a/deepteam/vulnerabilities/system_reconnaissance/system_reconnaissance.py b/deepteam/vulnerabilities/system_reconnaissance/system_reconnaissance.py
index 011e873e..3b382708 100644
--- a/deepteam/vulnerabilities/system_reconnaissance/system_reconnaissance.py
+++ b/deepteam/vulnerabilities/system_reconnaissance/system_reconnaissance.py
@@ -12,6 +12,7 @@ from deepteam.vulnerabilities.system_reconnaissance import (
)
from deepteam.vulnerabilities.utils import validate_vulnerability_types
from deepteam.metrics import SystemReconnaissanceMetric, BaseRedTeamingMetric
+from deepteam.metrics.types import EvaluationExample
from deepteam.attacks.multi_turn.types import CallbackType
from deepteam.test_case import RTTestCase
from deepteam.attacks.attack_simulator.schema import SyntheticDataList
@@ -40,6 +41,8 @@ class SystemReconnaissance(BaseVulnerability):
type.value for type in SystemReconnaissanceType
],
purpose: Optional[str] = None,
+ evaluation_examples: Optional[List[EvaluationExample]] = None,
+ evaluation_guidelines: Optional[List[str]] = None,
):
enum_types = validate_vulnerability_types(
self.get_name(), types=types, allowed_type=SystemReconnaissanceType
@@ -49,6 +52,8 @@ class SystemReconnaissance(BaseVulnerability):
self.simulator_model = simulator_model
self.evaluation_model = evaluation_model
self.purpose = purpose
+ self.evaluation_examples = evaluation_examples
+ self.evaluation_guidelines = evaluation_guidelines
super().__init__(types=enum_types)
def assess(
@@ -288,6 +293,8 @@ class SystemReconnaissance(BaseVulnerability):
model=self.evaluation_model,
async_mode=self.async_mode,
verbose_mode=self.verbose_mode,
+ evaluation_examples=self.evaluation_examples,
+ evaluation_guidelines=self.evaluation_guidelines,
)
def is_vulnerable(self) -> bool:
diff --git a/deepteam/vulnerabilities/tool_metadata_poisoning/tool_metadata_poisoning.py b/deepteam/vulnerabilities/tool_metadata_poisoning/tool_metadata_poisoning.py
index b7c0cb3e..4cd17f9f 100644
--- a/deepteam/vulnerabilities/tool_metadata_poisoning/tool_metadata_poisoning.py
+++ b/deepteam/vulnerabilities/tool_metadata_poisoning/tool_metadata_poisoning.py
@@ -13,6 +13,7 @@ from deepteam.vulnerabilities.tool_metadata_poisoning import (
)
from deepteam.vulnerabilities.utils import validate_vulnerability_types
from deepteam.metrics import ToolMetadataPoisoningMetric, BaseRedTeamingMetric
+from deepteam.metrics.types import EvaluationExample
from deepteam.attacks.multi_turn.types import CallbackType
from deepteam.test_case import RTTestCase
from deepteam.attacks.attack_simulator.schema import SyntheticDataList
@@ -44,6 +45,8 @@ class ToolMetadataPoisoning(BaseVulnerability):
type.value for type in ToolMetadataPoisoningType
],
purpose: Optional[str] = None,
+ evaluation_examples: Optional[List[EvaluationExample]] = None,
+ evaluation_guidelines: Optional[List[str]] = None,
):
enum_types = validate_vulnerability_types(
self.get_name(), types=types, allowed_type=ToolMetadataPoisoningType
@@ -53,6 +56,8 @@ class ToolMetadataPoisoning(BaseVulnerability):
self.simulator_model = simulator_model
self.evaluation_model = evaluation_model
self.purpose = purpose
+ self.evaluation_examples = evaluation_examples
+ self.evaluation_guidelines = evaluation_guidelines
super().__init__(types=enum_types)
def assess(
@@ -292,6 +297,8 @@ class ToolMetadataPoisoning(BaseVulnerability):
model=self.evaluation_model,
async_mode=self.async_mode,
verbose_mode=self.verbose_mode,
+ evaluation_examples=self.evaluation_examples,
+ evaluation_guidelines=self.evaluation_guidelines,
)
def is_vulnerable(self) -> bool:
diff --git a/deepteam/vulnerabilities/tool_orchestration_abuse/tool_orchestration_abuse.py b/deepteam/vulnerabilities/tool_orchestration_abuse/tool_orchestration_abuse.py
index ddf238d3..919bdb11 100644
--- a/deepteam/vulnerabilities/tool_orchestration_abuse/tool_orchestration_abuse.py
+++ b/deepteam/vulnerabilities/tool_orchestration_abuse/tool_orchestration_abuse.py
@@ -13,6 +13,7 @@ from deepteam.vulnerabilities.tool_orchestration_abuse import (
)
from deepteam.vulnerabilities.utils import validate_vulnerability_types
from deepteam.metrics import ToolOrchestrationMetric, BaseRedTeamingMetric
+from deepteam.metrics.types import EvaluationExample
from deepteam.attacks.multi_turn.types import CallbackType
from deepteam.test_case import RTTestCase
from deepteam.attacks.attack_simulator.schema import SyntheticDataList
@@ -44,6 +45,8 @@ class ToolOrchestrationAbuse(BaseVulnerability):
type.value for type in ToolOrchestrationAbuseType
],
purpose: Optional[str] = None,
+ evaluation_examples: Optional[List[EvaluationExample]] = None,
+ evaluation_guidelines: Optional[List[str]] = None,
):
enum_types = validate_vulnerability_types(
self.get_name(),
@@ -55,6 +58,8 @@ class ToolOrchestrationAbuse(BaseVulnerability):
self.simulator_model = simulator_model
self.evaluation_model = evaluation_model
self.purpose = purpose
+ self.evaluation_examples = evaluation_examples
+ self.evaluation_guidelines = evaluation_guidelines
super().__init__(types=enum_types)
def assess(
@@ -294,6 +299,8 @@ class ToolOrchestrationAbuse(BaseVulnerability):
model=self.evaluation_model,
async_mode=self.async_mode,
verbose_mode=self.verbose_mode,
+ evaluation_examples=self.evaluation_examples,
+ evaluation_guidelines=self.evaluation_guidelines,
)
def is_vulnerable(self) -> bool:
diff --git a/deepteam/vulnerabilities/toxicity/toxicity.py b/deepteam/vulnerabilities/toxicity/toxicity.py
index 8b05bd52..84eb796b 100644
--- a/deepteam/vulnerabilities/toxicity/toxicity.py
+++ b/deepteam/vulnerabilities/toxicity/toxicity.py
@@ -11,6 +11,7 @@ from deepteam.vulnerabilities import BaseVulnerability
from deepteam.vulnerabilities.toxicity import ToxicityType
from deepteam.vulnerabilities.utils import validate_vulnerability_types
from deepteam.metrics import ToxicityMetric, BaseRedTeamingMetric
+from deepteam.metrics.types import EvaluationExample
from deepteam.attacks.multi_turn.types import CallbackType
from deepteam.test_case import RTTestCase
from deepteam.attacks.attack_simulator.schema import SyntheticDataList
@@ -37,6 +38,8 @@ class Toxicity(BaseVulnerability):
type.value for type in ToxicityType
],
purpose: Optional[str] = None,
+ evaluation_examples: Optional[List[EvaluationExample]] = None,
+ evaluation_guidelines: Optional[List[str]] = None,
):
enum_types = validate_vulnerability_types(
self.get_name(), types=types, allowed_type=ToxicityType
@@ -46,6 +49,8 @@ class Toxicity(BaseVulnerability):
self.simulator_model = simulator_model
self.evaluation_model = evaluation_model
self.purpose = purpose
+ self.evaluation_examples = evaluation_examples
+ self.evaluation_guidelines = evaluation_guidelines
super().__init__(types=enum_types)
def assess(
@@ -281,6 +286,8 @@ class Toxicity(BaseVulnerability):
model=self.evaluation_model,
async_mode=self.async_mode,
verbose_mode=self.verbose_mode,
+ evaluation_examples=self.evaluation_examples,
+ evaluation_guidelines=self.evaluation_guidelines,
)
def is_vulnerable(self) -> bool:
diff --git a/deepteam/vulnerabilities/unexpected_code_execution/unexpected_code_execution.py b/deepteam/vulnerabilities/unexpected_code_execution/unexpected_code_execution.py
index aa4e29d7..ad7ca2f7 100644
--- a/deepteam/vulnerabilities/unexpected_code_execution/unexpected_code_execution.py
+++ b/deepteam/vulnerabilities/unexpected_code_execution/unexpected_code_execution.py
@@ -13,6 +13,7 @@ from deepteam.vulnerabilities.unexpected_code_execution import (
)
from deepteam.vulnerabilities.utils import validate_vulnerability_types
from deepteam.metrics import UnexpectedCodeExecutionMetric, BaseRedTeamingMetric
+from deepteam.metrics.types import EvaluationExample
from deepteam.attacks.multi_turn.types import CallbackType
from deepteam.test_case import RTTestCase
from deepteam.attacks.attack_simulator.schema import SyntheticDataList
@@ -41,6 +42,8 @@ class UnexpectedCodeExecution(BaseVulnerability):
type.value for type in UnexpectedCodeExecutionType
],
purpose: Optional[str] = None,
+ evaluation_examples: Optional[List[EvaluationExample]] = None,
+ evaluation_guidelines: Optional[List[str]] = None,
):
enum_types = validate_vulnerability_types(
self.get_name(),
@@ -52,6 +55,8 @@ class UnexpectedCodeExecution(BaseVulnerability):
self.simulator_model = simulator_model
self.evaluation_model = evaluation_model
self.purpose = purpose
+ self.evaluation_examples = evaluation_examples
+ self.evaluation_guidelines = evaluation_guidelines
super().__init__(types=enum_types)
def assess(
@@ -295,6 +300,8 @@ class UnexpectedCodeExecution(BaseVulnerability):
model=self.evaluation_model,
async_mode=self.async_mode,
verbose_mode=self.verbose_mode,
+ evaluation_examples=self.evaluation_examples,
+ evaluation_guidelines=self.evaluation_guidelines,
)
def is_vulnerable(self) -> bool: