Skip to content

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

backup before reset

penguine279fe623cf74
diff --git a/deepteam/attacks/attack_simulator/attack_simulator.py b/deepteam/attacks/attack_simulator/attack_simulator.py
index c681ea7c..62390880 100644
--- a/deepteam/attacks/attack_simulator/attack_simulator.py
+++ b/deepteam/attacks/attack_simulator/attack_simulator.py
@@ -1,6 +1,5 @@
import random
import asyncio
-from pydantic import BaseModel
from typing import List, Optional, Union
import inspect
@@ -264,25 +263,13 @@ class AttackSimulator:
test_case: RTTestCase,
ignore_errors: bool,
):
- from deepteam.attacks.multi_turn import (
- BadLikertJudge,
- CrescendoJailbreaking,
- LinearJailbreaking,
- SequentialJailbreak,
- TreeJailbreaking,
- )
from deepteam.test_case.test_case import RTTurn
-
- MULTI_TURN_ATTACKS = [
- BadLikertJudge,
- CrescendoJailbreaking,
- LinearJailbreaking,
- TreeJailbreaking,
- SequentialJailbreak,
- ]
+ from deepteam.attacks.multi_turn.base_multi_turn_attack import (
+ BaseMultiTurnAttack,
+ )
# Check if the attack is a multi-turn attack
- if type(attack) in MULTI_TURN_ATTACKS:
+ if isinstance(attack, BaseMultiTurnAttack):
# This is a multi-turn attack
attack_input = test_case.input
if attack_input is None:
@@ -292,11 +279,10 @@ class AttackSimulator:
turns = [RTTurn(role="user", content=attack_input)]
try:
- res = attack._get_turns(
+ res = attack.progress(
model_callback=self.model_callback,
turns=turns,
vulnerability=test_case.vulnerability,
- vulnerability_type=test_case.vulnerability_type.value,
)
test_case.turns = res
@@ -310,7 +296,7 @@ class AttackSimulator:
except Exception as e:
if ignore_errors:
test_case.error = (
- f"Error enhancing multi-turn attack: {str(e)}"
+ f"Error progressing multi-turn attack: {str(e)}"
)
return test_case
else:
@@ -373,41 +359,27 @@ class AttackSimulator:
test_case: RTTestCase,
ignore_errors: bool,
):
- from deepteam.attacks.multi_turn import (
- BadLikertJudge,
- CrescendoJailbreaking,
- LinearJailbreaking,
- SequentialJailbreak,
- TreeJailbreaking,
- )
from deepteam.test_case.test_case import RTTurn
+ from deepteam.attacks.multi_turn.base_multi_turn_attack import (
+ BaseMultiTurnAttack,
+ )
- MULTI_TURN_ATTACKS = [
- BadLikertJudge,
- CrescendoJailbreaking,
- LinearJailbreaking,
- TreeJailbreaking,
- SequentialJailbreak,
- ]
-
- if type(attack) in MULTI_TURN_ATTACKS:
+ if isinstance(attack, BaseMultiTurnAttack):
# This is multi-turn attack
attack_input = test_case.input
if attack_input is None:
return test_case
test_case.attack_method = attack.get_name()
- sig = inspect.signature(attack.a_enhance)
+ sig = inspect.signature(attack.a_progress)
turns = [RTTurn(role="user", content=attack_input)]
try:
- res = await attack._a_get_turns(
+ res = await attack.a_progress(
model_callback=self.model_callback,
turns=turns,
vulnerability=test_case.vulnerability,
- vulnerability_type=test_case.vulnerability_type.value,
)
-
test_case.turns = res
except ModelRefusalError as e:
@@ -418,7 +390,7 @@ class AttackSimulator:
raise
except:
if ignore_errors:
- test_case.error = "Error enhancing attack"
+ test_case.error = "Error progressing attack"
return test_case
else:
raise
diff --git a/deepteam/frameworks/frameworks.py b/deepteam/frameworks/frameworks.py
index 04beed14..1d501871 100644
--- a/deepteam/frameworks/frameworks.py
+++ b/deepteam/frameworks/frameworks.py
@@ -18,5 +18,8 @@ class AISafetyFramework:
attacks: Optional[List[BaseAttack]]
_has_dataset: bool = False
+ def get_name(self) -> str:
+ return self.name
+
class Config:
arbitrary_types_allowed = True
diff --git a/deepteam/frameworks/mitre/risk_categories.py b/deepteam/frameworks/mitre/risk_categories.py
index 3ea34230..3f0da8da 100644
--- a/deepteam/frameworks/mitre/risk_categories.py
+++ b/deepteam/frameworks/mitre/risk_categories.py
@@ -1,4 +1,4 @@
-from deepteam.frameworks.risk_category import RiskCategory
+from deepteam.risk_categories import RiskCategory
from deepteam.vulnerabilities import (
IllegalActivity,
PromptLeakage,
diff --git a/deepteam/frameworks/nist/risk_categories.py b/deepteam/frameworks/nist/risk_categories.py
index 45b04812..b6b25692 100644
--- a/deepteam/frameworks/nist/risk_categories.py
+++ b/deepteam/frameworks/nist/risk_categories.py
@@ -1,4 +1,4 @@
-from deepteam.frameworks.risk_category import RiskCategory
+from deepteam.risk_categories import RiskCategory
from deepteam.vulnerabilities import (
Bias,
Toxicity,
diff --git a/deepteam/frameworks/owasp/risk_categories.py b/deepteam/frameworks/owasp/risk_categories.py
index 0e339c45..deae4207 100644
--- a/deepteam/frameworks/owasp/risk_categories.py
+++ b/deepteam/frameworks/owasp/risk_categories.py
@@ -1,5 +1,5 @@
from typing import List
-from deepteam.frameworks.risk_category import RiskCategory
+from deepteam.risk_categories import RiskCategory
from deepteam.vulnerabilities import (
Bias,
Toxicity,
diff --git a/deepteam/red_teamer/red_teamer.py b/deepteam/red_teamer/red_teamer.py
index ab7a9e39..4242ed2b 100644
--- a/deepteam/red_teamer/red_teamer.py
+++ b/deepteam/red_teamer/red_teamer.py
@@ -19,7 +19,7 @@ from deepteam.test_case import RTTestCase
from deepeval.utils import get_or_create_event_loop
from deepteam.frameworks.frameworks import AISafetyFramework
-from deepteam.frameworks.risk_category import RiskCategory
+from deepteam.risk_categories import RiskCategory
from deepteam.telemetry import capture_red_teamer_run
from deepteam.attacks import BaseAttack
from deepteam.utils import (
@@ -37,7 +37,7 @@ from deepteam.red_teamer.risk_assessment import (
construct_risk_assessment_overview,
RiskAssessment,
)
-from deepteam.risks import getRiskCategory
+from deepteam.risk_categories.utils import getRiskCategory
from deepteam.confident.api import Api, HttpMethods, Endpoints
console = Console()
@@ -79,13 +79,13 @@ class RedTeamer:
ignore_errors: bool = False,
reuse_simulated_test_cases: bool = False,
metadata: Optional[dict] = None,
- ):
+ ) -> RiskAssessment:
if not framework or framework._has_dataset:
raise ValueError(
"Please pass in a valid framework that does not rely on a dataset."
)
- def assess_risk_category(category: RiskCategory):
+ def assess_risk_category(category: RiskCategory) -> RiskAssessment:
return self.red_team(
model_callback=model_callback,
attacks=category.attacks,
@@ -99,7 +99,6 @@ class RedTeamer:
_print_assessment=False,
)
- results = {}
progress = create_progress()
with progress:
task_id = add_pbar(
@@ -107,6 +106,8 @@ class RedTeamer:
description=f"⏳ Running red-teaming for {framework.get_name()} Framework",
total=len(framework.risk_categories),
)
+
+ # Assess each risk category and populate its test_cases
for risk_category in framework.risk_categories:
progress_2 = create_progress()
with progress_2:
@@ -115,12 +116,29 @@ class RedTeamer:
description=f"🖍️ Assessing risk-category: {risk_category.name}",
total=1,
)
- framework_assessment = assess_risk_category(risk_category)
- results[risk_category.name] = framework_assessment
+ assessment = assess_risk_category(risk_category)
+ risk_category.test_cases = assessment.test_cases
update_pbar(progress_2, risk_task_id, advance_to_end=True)
update_pbar(progress, task_id)
- return results
+ # Aggregate all test cases from all risk categories
+ all_test_cases = []
+ total_duration = 0
+ for category in framework.risk_categories:
+ if category.test_cases:
+ all_test_cases.extend(category.test_cases)
+
+ # Create the combined RiskAssessment
+ combined_assessment = RiskAssessment(
+ overview=construct_risk_assessment_overview(
+ red_teaming_test_cases=all_test_cases,
+ run_duration=total_duration,
+ ),
+ test_cases=all_test_cases,
+ risk_categories=framework.risk_categories,
+ )
+
+ return combined_assessment
async def _a_framework_assessment(
self,
@@ -132,13 +150,18 @@ class RedTeamer:
ignore_errors: bool = False,
reuse_simulated_test_cases: bool = False,
metadata: Optional[dict] = None,
- ) -> Dict[str, RiskAssessment]:
+ ) -> RiskAssessment:
if not framework or framework._has_dataset:
raise ValueError(
"Please pass in a valid framework that does not rely on a dataset."
)
- async def assess_risk_category(category: RiskCategory):
+ # Create a mapping of category to its assessment
+ category_to_task = {}
+
+ async def assess_risk_category(
+ category: RiskCategory,
+ ) -> RiskAssessment:
progress_2 = create_progress()
with progress_2:
risk_task_id = add_pbar(
@@ -163,28 +186,45 @@ class RedTeamer:
progress = create_progress()
with progress:
-
tasks = {
- asyncio.create_task(
- assess_risk_category(category)
- ): category.name
+ asyncio.create_task(assess_risk_category(category)): category
for category in framework.risk_categories
}
- results = {}
-
task_id = add_pbar(
progress,
description=f"⏳ Running red-teaming for {framework.get_name()} Framework",
total=len(framework.risk_categories),
)
+
+ # Collect assessments and populate risk category test_cases
for task in asyncio.as_completed(tasks):
- result = await task
- name = tasks[task]
- results[name] = result
+ assessment = await task
+ category = tasks[task]
+ category.test_cases = assessment.test_cases
update_pbar(progress, task_id)
- return results
+ # Aggregate all test cases from all risk categories
+ all_test_cases = []
+ total_duration = 0
+ for category in framework.risk_categories:
+ if category.test_cases:
+ all_test_cases.extend(category.test_cases)
+
+ # Calculate total duration from all categories
+ # (Note: we'll need to track this properly if needed)
+
+ # Create the combined RiskAssessment
+ combined_assessment = RiskAssessment(
+ overview=construct_risk_assessment_overview(
+ red_teaming_test_cases=all_test_cases,
+ run_duration=total_duration,
+ ),
+ test_cases=all_test_cases,
+ risk_categories=framework.risk_categories,
+ )
+
+ return combined_assessment
def red_team(
self,
@@ -211,7 +251,7 @@ class RedTeamer:
)
if framework and not framework._has_dataset:
- framework_assessment = self._framework_assessment(
+ risk_assessment = self._framework_assessment(
model_callback=model_callback,
simulator_model=simulator_model,
evaluation_model=evaluation_model,
@@ -221,10 +261,16 @@ class RedTeamer:
reuse_simulated_test_cases=reuse_simulated_test_cases,
metadata=metadata,
)
- self._print_framework_overview_table(
- framework_results=framework_assessment
- )
- return framework_assessment
+ self.risk_assessment = risk_assessment
+ self.test_cases = risk_assessment.test_cases
+
+ # Print the framework overview table using risk categories
+ if risk_assessment.risk_categories:
+ self._print_framework_overview_table(
+ risk_categories=risk_assessment.risk_categories
+ )
+
+ return risk_assessment
if self.async_mode:
validate_model_callback_signature(
@@ -379,11 +425,10 @@ class RedTeamer:
test_cases=red_teaming_test_cases,
)
self.test_cases = red_teaming_test_cases
- self._print_risk_assessment()
- self._post_risk_assessment()
if _print_assessment:
- self._print_risk_assessment(self.risk_assessment)
+ self._print_risk_assessment()
+ self._post_risk_assessment()
return self.risk_assessment
@@ -414,7 +459,7 @@ class RedTeamer:
if framework and not framework._has_dataset:
loop = get_or_create_event_loop()
- framework_assessment = loop.run_until_complete(
+ risk_assessment = loop.run_until_complete(
await self._a_framework_assessment(
model_callback=model_callback,
simulator_model=simulator_model,
@@ -426,10 +471,16 @@ class RedTeamer:
metadata=metadata,
)
)
- self._print_framework_overview_table(
- framework_results=framework_assessment
- )
- return framework_assessment
+ self.risk_assessment = risk_assessment
+ self.test_cases = risk_assessment.test_cases
+
+ # Print the framework overview table using risk categories
+ if risk_assessment.risk_categories:
+ self._print_framework_overview_table(
+ risk_categories=risk_assessment.risk_categories
+ )
+
+ return risk_assessment
if framework:
if framework._has_dataset:
@@ -568,11 +619,10 @@ class RedTeamer:
test_cases=red_teaming_test_cases,
)
self.test_cases = red_teaming_test_cases
- self._print_risk_assessment()
- self._post_risk_assessment()
if _print_assessment:
- self._print_risk_assessment(self.risk_assessment)
+ self._print_risk_assessment()
+ self._post_risk_assessment()
return self.risk_assessment
@@ -771,9 +821,13 @@ class RedTeamer:
)
return red_teaming_test_cases
- def _print_risk_assessment(self, risk_assessment=None):
+ def _print_risk_assessment(
+ self, risk_assessment: Optional[RiskAssessment] = None
+ ):
if risk_assessment is None:
- return
+ risk_assessment = self.risk_assessment
+ if risk_assessment is None:
+ return
console = Console()
@@ -953,13 +1007,21 @@ class RedTeamer:
f"[link={link}]{link}[/link]"
)
webbrowser.open(link)
-
-
- def _print_framework_overview_table(self, framework_results: dict):
- for category_name in sorted(framework_results.keys()):
- assessment = framework_results[category_name]
- self._print_risk_assessment(assessment)
+ def _print_framework_overview_table(
+ self, risk_categories: List[RiskCategory]
+ ):
+ # Print individual risk assessments for each category
+ for category in sorted(risk_categories, key=lambda c: c.name):
+ if category.test_cases:
+ assessment = RiskAssessment(
+ overview=construct_risk_assessment_overview(
+ red_teaming_test_cases=category.test_cases,
+ run_duration=0,
+ ),
+ test_cases=category.test_cases,
+ )
+ self._print_risk_assessment(assessment)
console = Console()
@@ -989,30 +1051,32 @@ class RedTeamer:
table.add_column("Vulnerabilities Tested", style="white", width=25)
table.add_column("Attack Methods Used", style="white", width=25)
- for category_name in sorted(framework_results.keys()):
- assessment = framework_results[category_name]
+ for category in sorted(risk_categories, key=lambda c: c.name):
+ if not category.test_cases:
+ continue
- overview = assessment.overview
+ # Calculate pass/fail/error stats from test cases
passing = sum(
- result.passing for result in overview.vulnerability_type_results
- )
- failing = sum(
- result.failing for result in overview.vulnerability_type_results
+ 1
+ for tc in category.test_cases
+ if tc.score is not None and tc.score > 0
)
errored = sum(
- result.errored for result in overview.vulnerability_type_results
+ 1 for tc in category.test_cases if tc.error is not None
)
+ failing = len(category.test_cases) - passing - errored
total = passing + failing
pass_rate = passing / total if total > 0 else 0.0
+ # Get vulnerabilities directly from risk_category
+ vuln_lines = []
vulnerability_groups = defaultdict(list)
- for result in overview.vulnerability_type_results:
- vulnerability_groups[result.vulnerability].append(
- result.vulnerability_type.value
- )
+ for vuln in category.vulnerabilities:
+ vuln_name = vuln.get_name()
+ for vuln_type in vuln.get_types():
+ vulnerability_groups[vuln_name].append(vuln_type.value)
- vuln_lines = []
for (
vulnerability_name,
vulnerability_types,
@@ -1023,10 +1087,8 @@ class RedTeamer:
vulnerabilty_names = "\n".join(vuln_lines) if vuln_lines else "N/A"
- # ----- Attack methods (simple list) -----
- attack_list = [
- amr.attack_method for amr in overview.attack_method_results
- ]
+ # Get attacks directly from risk_category
+ attack_list = [attack.get_name() for attack in category.attacks]
attack_names = "\n".join(attack_list) if attack_list else "N/A"
# Color-coded pass rate
@@ -1038,7 +1100,7 @@ class RedTeamer:
pass_rate_str = f"[bold red]{pass_rate:.0%}[/bold red]"
table.add_row(
- category_name,
+ category.name,
pass_rate_str,
str(passing),
str(failing),
diff --git a/deepteam/red_teamer/risk_assessment.py b/deepteam/red_teamer/risk_assessment.py
index c7ce2058..78655e54 100644
--- a/deepteam/red_teamer/risk_assessment.py
+++ b/deepteam/red_teamer/risk_assessment.py
@@ -1,5 +1,5 @@
-from pydantic import BaseModel, Field
-from typing import Dict, Optional, List, Any, Union
+from pydantic import BaseModel, Field, ConfigDict
+from typing import Dict, Optional, List, Any, Union, TYPE_CHECKING
import datetime
import os
import json
@@ -7,6 +7,7 @@ from enum import Enum
from deepteam.vulnerabilities.types import VulnerabilityType
from deepteam.test_case import RTTestCase
+from deepteam.risk_categories.risk_category import RiskCategory
class TestCasesList(list):
@@ -125,8 +126,10 @@ class EnumEncoder(json.JSONEncoder):
class RiskAssessment(BaseModel):
+ model_config = ConfigDict(arbitrary_types_allowed=True)
overview: RedTeamingOverview
test_cases: List[RTTestCase]
+ risk_categories: Optional[List[RiskCategory]] = Field(None)
def __init__(self, **data):
super().__init__(**data)
@@ -255,3 +258,6 @@ def construct_risk_assessment_overview(
errored=errored,
run_duration=run_duration,
)
+
+
+RiskAssessment.model_rebuild()
diff --git a/deepteam/risk_categories/__init__.py b/deepteam/risk_categories/__init__.py
new file mode 100644
index 00000000..f0167510
--- /dev/null
+++ b/deepteam/risk_categories/__init__.py
@@ -0,0 +1,4 @@
+from .utils import getRiskCategory
+from .risk_category import RiskCategory
+
+__all__ = ["getRiskCategory", "RiskCategory"]
diff --git a/deepteam/frameworks/risk_category.py b/deepteam/risk_categories/risk_category.py
similarity index 75%
rename from deepteam/frameworks/risk_category.py
rename to deepteam/risk_categories/risk_category.py
index b42ba0e3..f05718a1 100644
--- a/deepteam/frameworks/risk_category.py
+++ b/deepteam/risk_categories/risk_category.py
@@ -1,9 +1,11 @@
from typing import List, Optional
+from dataclasses import dataclass
+
from deepteam.vulnerabilities import (
BaseVulnerability,
)
from deepteam.attacks import BaseAttack
-from dataclasses import dataclass
+from deepteam.test_case import RTTestCase
@dataclass
@@ -11,3 +13,4 @@ class RiskCategory:
name: str
vulnerabilities: List[BaseVulnerability]
attacks: List[BaseAttack]
+ test_cases: Optional[List[RTTestCase]] = None
diff --git a/deepteam/risks/risks.py b/deepteam/risk_categories/utils.py
similarity index 100%
rename from deepteam/risks/risks.py
rename to deepteam/risk_categories/utils.py
diff --git a/deepteam/risks/__init__.py b/deepteam/risks/__init__.py
deleted file mode 100644
index 9b0d31eb..00000000
--- a/deepteam/risks/__init__.py
+++ /dev/null
@@ -1 +0,0 @@
-from .risks import getRiskCategory
diff --git a/deepteam/vulnerabilities/bfla/bfla.py b/deepteam/vulnerabilities/bfla/bfla.py
index f735b07d..728af4f3 100644
--- a/deepteam/vulnerabilities/bfla/bfla.py
+++ b/deepteam/vulnerabilities/bfla/bfla.py
@@ -14,7 +14,7 @@ from deepteam.metrics import BFLAMetric, BaseRedTeamingMetric
from deepteam.attacks.multi_turn.types import CallbackType
from deepteam.test_case import RTTestCase
from deepteam.attacks.attack_simulator.schema import SyntheticDataList
-from deepteam.risks import getRiskCategory
+from deepteam.risk_categories.utils import getRiskCategory
from .template import BFLATemplate
BFLALiteral = Literal[
diff --git a/deepteam/vulnerabilities/bias/bias.py b/deepteam/vulnerabilities/bias/bias.py
index 612532d8..ccca3536 100644
--- a/deepteam/vulnerabilities/bias/bias.py
+++ b/deepteam/vulnerabilities/bias/bias.py
@@ -13,7 +13,7 @@ from deepteam.metrics import BiasMetric, BaseRedTeamingMetric
from deepteam.attacks.multi_turn.types import CallbackType
from deepteam.test_case import RTTestCase
from deepteam.attacks.attack_simulator.schema import SyntheticDataList
-from deepteam.risks import getRiskCategory
+from deepteam.risk_categories.utils import getRiskCategory
from .template import BiasTemplate
BiasLiteralType = Literal["religion", "politics", "gender", "race"]
diff --git a/deepteam/vulnerabilities/bola/bola.py b/deepteam/vulnerabilities/bola/bola.py
index 7354505f..4277d7c4 100644
--- a/deepteam/vulnerabilities/bola/bola.py
+++ b/deepteam/vulnerabilities/bola/bola.py
@@ -14,7 +14,7 @@ from deepteam.metrics import BOLAMetric, BaseRedTeamingMetric
from deepteam.attacks.multi_turn.types import CallbackType
from deepteam.test_case import RTTestCase
from deepteam.attacks.attack_simulator.schema import SyntheticDataList
-from deepteam.risks import getRiskCategory
+from deepteam.risk_categories.utils import getRiskCategory
from .template import BOLATemplate
BOLALiteral = Literal[
diff --git a/deepteam/vulnerabilities/child_protection/child_protection.py b/deepteam/vulnerabilities/child_protection/child_protection.py
index 1f0aad48..c4e27749 100644
--- a/deepteam/vulnerabilities/child_protection/child_protection.py
+++ b/deepteam/vulnerabilities/child_protection/child_protection.py
@@ -13,7 +13,7 @@ from deepteam.metrics import ChildProtectionMetric, BaseRedTeamingMetric
from deepteam.attacks.multi_turn.types import CallbackType
from deepteam.test_case import RTTestCase
from deepteam.attacks.attack_simulator.schema import SyntheticDataList
-from deepteam.risks import getRiskCategory
+from deepteam.risk_categories.utils import getRiskCategory
from .template import ChildProtectionTemplate
ChildProtectionLiteralType = Literal[
diff --git a/deepteam/vulnerabilities/competition/competition.py b/deepteam/vulnerabilities/competition/competition.py
index cf699bd7..bb1acbff 100644
--- a/deepteam/vulnerabilities/competition/competition.py
+++ b/deepteam/vulnerabilities/competition/competition.py
@@ -15,7 +15,7 @@ from deepteam.metrics import CompetitorsMetric, BaseRedTeamingMetric
from deepteam.attacks.multi_turn.types import CallbackType
from deepteam.test_case import RTTestCase
from deepteam.attacks.attack_simulator.schema import SyntheticDataList
-from deepteam.risks import getRiskCategory
+from deepteam.risk_categories.utils import getRiskCategory
from .template import CompetitionTemplate
diff --git a/deepteam/vulnerabilities/custom/custom.py b/deepteam/vulnerabilities/custom/custom.py
index ad0f4ae9..13c518f5 100644
--- a/deepteam/vulnerabilities/custom/custom.py
+++ b/deepteam/vulnerabilities/custom/custom.py
@@ -12,7 +12,7 @@ from deepteam.vulnerabilities import BaseVulnerability
from deepteam.metrics import BaseRedTeamingMetric, HarmMetric
from deepteam.attacks.multi_turn.types import CallbackType
from deepteam.attacks.attack_simulator.schema import SyntheticDataList
-from deepteam.risks import getRiskCategory
+from deepteam.risk_categories.utils import getRiskCategory
from deepteam.test_case import RTTestCase
from .template import CustomVulnerabilityTemplate
diff --git a/deepteam/vulnerabilities/debug_access/debug_access.py b/deepteam/vulnerabilities/debug_access/debug_access.py
index 0cc54998..66d91f7f 100644
--- a/deepteam/vulnerabilities/debug_access/debug_access.py
+++ b/deepteam/vulnerabilities/debug_access/debug_access.py
@@ -14,7 +14,7 @@ from deepteam.metrics import DebugAccessMetric, BaseRedTeamingMetric
from deepteam.attacks.multi_turn.types import CallbackType
from deepteam.test_case import RTTestCase
from deepteam.attacks.attack_simulator.schema import SyntheticDataList
-from deepteam.risks import getRiskCategory
+from deepteam.risk_categories.utils import getRiskCategory
from .template import DebugAccessTemplate
DebugAccessLiteral = Literal[
diff --git a/deepteam/vulnerabilities/ethics/ethics.py b/deepteam/vulnerabilities/ethics/ethics.py
index 8f16d78c..76ca2e03 100644
--- a/deepteam/vulnerabilities/ethics/ethics.py
+++ b/deepteam/vulnerabilities/ethics/ethics.py
@@ -13,7 +13,7 @@ from deepteam.metrics import EthicsMetric, BaseRedTeamingMetric
from deepteam.attacks.multi_turn.types import CallbackType
from deepteam.test_case import RTTestCase
from deepteam.attacks.attack_simulator.schema import SyntheticDataList
-from deepteam.risks import getRiskCategory
+from deepteam.risk_categories.utils import getRiskCategory
from .template import EthicsTemplate
EthicsLiteralType = Literal[
diff --git a/deepteam/vulnerabilities/excessive_agency/excessive_agency.py b/deepteam/vulnerabilities/excessive_agency/excessive_agency.py
index eba5481e..c93f0bb3 100644
--- a/deepteam/vulnerabilities/excessive_agency/excessive_agency.py
+++ b/deepteam/vulnerabilities/excessive_agency/excessive_agency.py
@@ -53,7 +53,7 @@ class ExcessiveAgency(BaseVulnerability):
model_callback: CallbackType,
purpose: Optional[str] = None,
) -> Dict[ExcessiveAgencyType, List[RTTestCase]]:
- from deepteam.risks import getRiskCategory
+ from deepteam.risk_categories.utils import getRiskCategory
validate_model_callback_signature(
model_callback=model_callback,
@@ -108,7 +108,7 @@ class ExcessiveAgency(BaseVulnerability):
model_callback: CallbackType,
purpose: Optional[str] = None,
) -> Dict[ExcessiveAgencyType, List[RTTestCase]]:
- from deepteam.risks import getRiskCategory
+ from deepteam.risk_categories.utils import getRiskCategory
validate_model_callback_signature(
model_callback=model_callback,
diff --git a/deepteam/vulnerabilities/fairness/fairness.py b/deepteam/vulnerabilities/fairness/fairness.py
index 1d76e1c6..d80dae72 100644
--- a/deepteam/vulnerabilities/fairness/fairness.py
+++ b/deepteam/vulnerabilities/fairness/fairness.py
@@ -13,7 +13,7 @@ from deepteam.metrics import FairnessMetric, BaseRedTeamingMetric
from deepteam.attacks.multi_turn.types import CallbackType
from deepteam.test_case import RTTestCase
from deepteam.attacks.attack_simulator.schema import SyntheticDataList
-from deepteam.risks import getRiskCategory
+from deepteam.risk_categories.utils import getRiskCategory
from .template import FairnessTemplate
FairnessLiteralType = Literal[
diff --git a/deepteam/vulnerabilities/goal_theft/goal_theft.py b/deepteam/vulnerabilities/goal_theft/goal_theft.py
index 2582b178..7c8a5f0f 100644
--- a/deepteam/vulnerabilities/goal_theft/goal_theft.py
+++ b/deepteam/vulnerabilities/goal_theft/goal_theft.py
@@ -55,7 +55,7 @@ class GoalTheft(BaseVulnerability):
model_callback: CallbackType,
purpose: Optional[str] = None,
) -> Dict[GoalTheftType, List[RTTestCase]]:
- from deepteam.risks import getRiskCategory
+ from deepteam.risk_categories.utils import getRiskCategory
validate_model_callback_signature(
model_callback=model_callback,
@@ -110,7 +110,7 @@ class GoalTheft(BaseVulnerability):
model_callback: CallbackType,
purpose: Optional[str] = None,
) -> Dict[GoalTheftType, List[RTTestCase]]:
- from deepteam.risks import getRiskCategory
+ from deepteam.risk_categories.utils import getRiskCategory
validate_model_callback_signature(
model_callback=model_callback,
diff --git a/deepteam/vulnerabilities/graphic_content/graphic_content.py b/deepteam/vulnerabilities/graphic_content/graphic_content.py
index 9ee5a3aa..606db00b 100644
--- a/deepteam/vulnerabilities/graphic_content/graphic_content.py
+++ b/deepteam/vulnerabilities/graphic_content/graphic_content.py
@@ -14,7 +14,7 @@ from deepteam.metrics import GraphicMetric, BaseRedTeamingMetric
from deepteam.attacks.multi_turn.types import CallbackType
from deepteam.test_case import RTTestCase
from deepteam.attacks.attack_simulator.schema import SyntheticDataList
-from deepteam.risks import getRiskCategory
+from deepteam.risk_categories.utils import getRiskCategory
from .template import GraphicContentTemplate
GraphicContentLiteral = Literal[
diff --git a/deepteam/vulnerabilities/illegal_activity/illegal_activity.py b/deepteam/vulnerabilities/illegal_activity/illegal_activity.py
index e28cc246..ab6af4e2 100644
--- a/deepteam/vulnerabilities/illegal_activity/illegal_activity.py
+++ b/deepteam/vulnerabilities/illegal_activity/illegal_activity.py
@@ -14,7 +14,7 @@ from deepteam.metrics import IllegalMetric, BaseRedTeamingMetric
from deepteam.attacks.multi_turn.types import CallbackType
from deepteam.test_case import RTTestCase
from deepteam.attacks.attack_simulator.schema import SyntheticDataList
-from deepteam.risks import getRiskCategory
+from deepteam.risk_categories.utils import getRiskCategory
from .template import IllegalActivityTemplate
IllegalActivityLiteral = Literal[
diff --git a/deepteam/vulnerabilities/intellectual_property/intellectual_property.py b/deepteam/vulnerabilities/intellectual_property/intellectual_property.py
index 6f34c29f..3b7bd01c 100644
--- a/deepteam/vulnerabilities/intellectual_property/intellectual_property.py
+++ b/deepteam/vulnerabilities/intellectual_property/intellectual_property.py
@@ -16,7 +16,7 @@ from deepteam.metrics import IntellectualPropertyMetric, BaseRedTeamingMetric
from deepteam.attacks.multi_turn.types import CallbackType
from deepteam.test_case import RTTestCase
from deepteam.attacks.attack_simulator.schema import SyntheticDataList
-from deepteam.risks import getRiskCategory
+from deepteam.risk_categories.utils import getRiskCategory
from .template import IntellectualPropertyTemplate
IntellectualPropertyLiteral = Literal[
diff --git a/deepteam/vulnerabilities/misinformation/misinformation.py b/deepteam/vulnerabilities/misinformation/misinformation.py
index 12698444..32ad0f92 100644
--- a/deepteam/vulnerabilities/misinformation/misinformation.py
+++ b/deepteam/vulnerabilities/misinformation/misinformation.py
@@ -14,7 +14,7 @@ from deepteam.metrics import MisinformationMetric, BaseRedTeamingMetric
from deepteam.attacks.multi_turn.types import CallbackType
from deepteam.test_case import RTTestCase
from deepteam.attacks.attack_simulator.schema import SyntheticDataList
-from deepteam.risks import getRiskCategory
+from deepteam.risk_categories.utils import getRiskCategory
from .template import MisinformationTemplate
diff --git a/deepteam/vulnerabilities/personal_safety/personal_safety.py b/deepteam/vulnerabilities/personal_safety/personal_safety.py
index 8be4d671..ab3aed53 100644
--- a/deepteam/vulnerabilities/personal_safety/personal_safety.py
+++ b/deepteam/vulnerabilities/personal_safety/personal_safety.py
@@ -14,7 +14,7 @@ from deepteam.metrics import SafetyMetric, BaseRedTeamingMetric
from deepteam.attacks.multi_turn.types import CallbackType
from deepteam.test_case import RTTestCase
from deepteam.attacks.attack_simulator.schema import SyntheticDataList
-from deepteam.risks import getRiskCategory
+from deepteam.risk_categories.utils import getRiskCategory
from .template import PersonalSafetyTemplate
PersonalSafetyLiteral = Literal[
diff --git a/deepteam/vulnerabilities/pii_leakage/pii_leakage.py b/deepteam/vulnerabilities/pii_leakage/pii_leakage.py
index 520097f4..28bd6b76 100644
--- a/deepteam/vulnerabilities/pii_leakage/pii_leakage.py
+++ b/deepteam/vulnerabilities/pii_leakage/pii_leakage.py
@@ -14,7 +14,7 @@ from deepteam.metrics import PIIMetric, BaseRedTeamingMetric
from deepteam.attacks.multi_turn.types import CallbackType
from deepteam.test_case import RTTestCase
from deepteam.attacks.attack_simulator.schema import SyntheticDataList
-from deepteam.risks import getRiskCategory
+from deepteam.risk_categories.utils import getRiskCategory
from .template import PIILeakageTemplate
PIILeakageLiteral = Literal[
diff --git a/deepteam/vulnerabilities/prompt_leakage/prompt_leakage.py b/deepteam/vulnerabilities/prompt_leakage/prompt_leakage.py
index 8bf44b2e..d225dc6f 100644
--- a/deepteam/vulnerabilities/prompt_leakage/prompt_leakage.py
+++ b/deepteam/vulnerabilities/prompt_leakage/prompt_leakage.py
@@ -14,7 +14,7 @@ from deepteam.metrics import PromptExtractionMetric, BaseRedTeamingMetric
from deepteam.attacks.multi_turn.types import CallbackType
from deepteam.test_case import RTTestCase
from deepteam.attacks.attack_simulator.schema import SyntheticDataList
-from deepteam.risks import getRiskCategory
+from deepteam.risk_categories.utils import getRiskCategory
from .template import PromptLeakageTemplate
PromptLeakageLiteral = Literal[
diff --git a/deepteam/vulnerabilities/rbac/rbac.py b/deepteam/vulnerabilities/rbac/rbac.py
index 233983de..cd977454 100644
--- a/deepteam/vulnerabilities/rbac/rbac.py
+++ b/deepteam/vulnerabilities/rbac/rbac.py
@@ -14,7 +14,7 @@ from deepteam.metrics import RBACMetric, BaseRedTeamingMetric
from deepteam.attacks.multi_turn.types import CallbackType
from deepteam.test_case import RTTestCase
from deepteam.attacks.attack_simulator.schema import SyntheticDataList
-from deepteam.risks import getRiskCategory
+from deepteam.risk_categories.utils import getRiskCategory
from .template import RBACTemplate
RBACLiteral = Literal[
diff --git a/deepteam/vulnerabilities/recursive_hijacking/recursive_hijacking.py b/deepteam/vulnerabilities/recursive_hijacking/recursive_hijacking.py
index 000db93a..f03fa382 100644
--- a/deepteam/vulnerabilities/recursive_hijacking/recursive_hijacking.py
+++ b/deepteam/vulnerabilities/recursive_hijacking/recursive_hijacking.py
@@ -59,7 +59,7 @@ class RecursiveHijacking(BaseVulnerability):
model_callback: CallbackType,
purpose: Optional[str] = None,
) -> Dict[RecursiveHijackingType, List[RTTestCase]]:
- from deepteam.risks import getRiskCategory
+ from deepteam.risk_categories.utils import getRiskCategory
validate_model_callback_signature(
model_callback=model_callback,
@@ -114,7 +114,7 @@ class RecursiveHijacking(BaseVulnerability):
model_callback: CallbackType,
purpose: Optional[str] = None,
) -> Dict[RecursiveHijackingType, List[RTTestCase]]:
- from deepteam.risks import getRiskCategory
+ from deepteam.risk_categories.utils import getRiskCategory
validate_model_callback_signature(
model_callback=model_callback,
diff --git a/deepteam/vulnerabilities/robustness/robustness.py b/deepteam/vulnerabilities/robustness/robustness.py
index 5095f923..fe2412f0 100644
--- a/deepteam/vulnerabilities/robustness/robustness.py
+++ b/deepteam/vulnerabilities/robustness/robustness.py
@@ -54,7 +54,7 @@ class Robustness(BaseVulnerability):
model_callback: CallbackType,
purpose: Optional[str] = None,
) -> Dict[RobustnessType, List[RTTestCase]]:
- from deepteam.risks import getRiskCategory
+ from deepteam.risk_categories.utils import getRiskCategory
validate_model_callback_signature(
model_callback=model_callback,
@@ -109,7 +109,7 @@ class Robustness(BaseVulnerability):
model_callback: CallbackType,
purpose: Optional[str] = None,
) -> Dict[RobustnessType, List[RTTestCase]]:
- from deepteam.risks import getRiskCategory
+ from deepteam.risk_categories.utils import getRiskCategory
validate_model_callback_signature(
model_callback=model_callback,
diff --git a/deepteam/vulnerabilities/shell_injection/shell_injection.py b/deepteam/vulnerabilities/shell_injection/shell_injection.py
index 3e1386b9..1648faf0 100644
--- a/deepteam/vulnerabilities/shell_injection/shell_injection.py
+++ b/deepteam/vulnerabilities/shell_injection/shell_injection.py
@@ -14,7 +14,7 @@ from deepteam.metrics import ShellInjectionMetric, BaseRedTeamingMetric
from deepteam.attacks.multi_turn.types import CallbackType
from deepteam.test_case import RTTestCase
from deepteam.attacks.attack_simulator.schema import SyntheticDataList
-from deepteam.risks import getRiskCategory
+from deepteam.risk_categories.utils import getRiskCategory
from .template import ShellInjectionTemplate
ShellInjectionLiteral = Literal[
diff --git a/deepteam/vulnerabilities/sql_injection/sql_injection.py b/deepteam/vulnerabilities/sql_injection/sql_injection.py
index e888f55e..08436c5f 100644
--- a/deepteam/vulnerabilities/sql_injection/sql_injection.py
+++ b/deepteam/vulnerabilities/sql_injection/sql_injection.py
@@ -14,7 +14,7 @@ from deepteam.metrics import SQLInjectionMetric, BaseRedTeamingMetric
from deepteam.attacks.multi_turn.types import CallbackType
from deepteam.test_case import RTTestCase
from deepteam.attacks.attack_simulator.schema import SyntheticDataList
-from deepteam.risks import getRiskCategory
+from deepteam.risk_categories.utils import getRiskCategory
from .template import SQLInjectionTemplate
SQLInjectionLiteral = Literal[
diff --git a/deepteam/vulnerabilities/ssrf/ssrf.py b/deepteam/vulnerabilities/ssrf/ssrf.py
index e80730ac..e5cdea99 100644
--- a/deepteam/vulnerabilities/ssrf/ssrf.py
+++ b/deepteam/vulnerabilities/ssrf/ssrf.py
@@ -14,7 +14,7 @@ from deepteam.metrics import SSRFMetric, BaseRedTeamingMetric
from deepteam.attacks.multi_turn.types import CallbackType
from deepteam.test_case import RTTestCase
from deepteam.attacks.attack_simulator.schema import SyntheticDataList
-from deepteam.risks import getRiskCategory
+from deepteam.risk_categories.utils import getRiskCategory
from .template import SSRFTemplate
SSRFLiteral = Literal[
diff --git a/deepteam/vulnerabilities/toxicity/toxicity.py b/deepteam/vulnerabilities/toxicity/toxicity.py
index ef5464bf..192ec958 100644
--- a/deepteam/vulnerabilities/toxicity/toxicity.py
+++ b/deepteam/vulnerabilities/toxicity/toxicity.py
@@ -14,7 +14,7 @@ from deepteam.metrics import ToxicityMetric, BaseRedTeamingMetric
from deepteam.attacks.multi_turn.types import CallbackType
from deepteam.test_case import RTTestCase
from deepteam.attacks.attack_simulator.schema import SyntheticDataList
-from deepteam.risks import getRiskCategory
+from deepteam.risk_categories.utils import getRiskCategory
from .template import ToxicityTemplate
ToxicityLiteral = Literal["profanity", "insults", "threats", "mockery"]
diff --git a/kk.py b/kk.py
index 994bab52..9c2bc0e6 100644
--- a/kk.py
+++ b/kk.py
@@ -2,6 +2,7 @@ from deepteam import red_team
from deepteam.vulnerabilities import Bias, BOLA
from deepteam.attacks.single_turn import Roleplay
import os
+from deepteam.frameworks import OWASPTop10
async def your_callback(input: str) -> str:
@@ -11,9 +12,10 @@ async def your_callback(input: str) -> str:
print(os.environ.get("CONFIDENT_API_KEY"))
risk_assessment = red_team(
- attacks=[Roleplay()],
- vulnerabilities=[Bias(), BOLA()],
+ # attacks=[Roleplay()],
+ # vulnerabilities=[Bias(), BOLA()],
model_callback=your_callback,
+ framework=OWASPTop10(),
)
from deepteam.vulnerabilities.constants import (
diff --git a/tests/test_frameworks/test_mitre.py b/tests/test_frameworks/test_mitre.py
index ebefc393..f46bae89 100644
--- a/tests/test_frameworks/test_mitre.py
+++ b/tests/test_frameworks/test_mitre.py
@@ -2,7 +2,7 @@ import pytest
from deepteam.frameworks import MITRE
from deepteam.vulnerabilities import BaseVulnerability
from deepteam.attacks import BaseAttack
-from deepteam.frameworks.risk_category import RiskCategory
+from deepteam.risk_categories import RiskCategory
from deepteam.frameworks.mitre.risk_categories import MITRE_CATEGORIES
from deepteam import red_team
diff --git a/tests/test_frameworks/test_nist.py b/tests/test_frameworks/test_nist.py
index 833efe99..44f1301a 100644
--- a/tests/test_frameworks/test_nist.py
+++ b/tests/test_frameworks/test_nist.py
@@ -3,7 +3,7 @@ from deepteam.frameworks import NIST
from deepteam.vulnerabilities import BaseVulnerability
from deepteam.attacks import BaseAttack
from deepteam.frameworks.nist.risk_categories import NIST_CATEGORIES
-from deepteam.frameworks.risk_category import RiskCategory
+from deepteam.risk_categories import RiskCategory
from deepteam import red_team
diff --git a/tests/test_frameworks/test_owasp_top_10.py b/tests/test_frameworks/test_owasp_top_10.py
index 1b98f4ff..18e1603e 100644
--- a/tests/test_frameworks/test_owasp_top_10.py
+++ b/tests/test_frameworks/test_owasp_top_10.py
@@ -3,7 +3,7 @@ from deepteam.frameworks import OWASPTop10
from deepteam.vulnerabilities import BaseVulnerability
from deepteam.attacks import BaseAttack
from deepteam.frameworks.owasp.risk_categories import OWASP_CATEGORIES
-from deepteam.frameworks.risk_category import RiskCategory
+from deepteam.risk_categories import RiskCategory
from deepteam import red_team