DeepTeam is a framework to red team LLMs and AI agents.
reformat
penguine035762f8c066
diff --git a/deepteam/attacks/attack_simulator/attack_simulator.py b/deepteam/attacks/attack_simulator/attack_simulator.pyindex 827b1f31..81fedbbb 100644--- a/deepteam/attacks/attack_simulator/attack_simulator.py+++ b/deepteam/attacks/attack_simulator/attack_simulator.py@@ -21,7 +21,6 @@ from deepteam.attacks.attack_simulator.schema import SyntheticDataListfrom deepteam.errors import ModelRefusalError-class SimulatedAttack(BaseModel):vulnerability: strvulnerability_type: Union[Enum, VulnerabilityType]diff --git a/deepteam/red_teamer/red_teamer.py b/deepteam/red_teamer/red_teamer.pyindex c7b24cc1..6db4f39f 100644--- a/deepteam/red_teamer/red_teamer.py+++ b/deepteam/red_teamer/red_teamer.py@@ -362,7 +362,6 @@ class RedTeamer:if simulated_attack.error is not None:red_teaming_test_case.error = simulated_attack.error-for _vulnerability in vulnerabilities:if vulnerability_type in _vulnerability.types:metric: BaseRedTeamingMetric = _vulnerability._get_metric(diff --git a/deepteam/vulnerabilities/agentic/excessive_agency/template.py b/deepteam/vulnerabilities/agentic/excessive_agency/template.pyindex 478bff1d..de246d1e 100644--- a/deepteam/vulnerabilities/agentic/excessive_agency/template.py+++ b/deepteam/vulnerabilities/agentic/excessive_agency/template.py@@ -77,7 +77,8 @@ class ExcessiveAgencyTemplate:@staticmethoddef generate_permissions_baseline_attacks(- max_goldens: int, purpose: Optional[str] = None,+ max_goldens: int,+ purpose: Optional[str] = None,) -> str:return f"""@@ -197,7 +198,8 @@ class ExcessiveAgencyTemplate:@staticmethoddef generate_autonomy_baseline_attacks(- max_goldens: int, purpose: Optional[str] = None,+ max_goldens: int,+ purpose: Optional[str] = None,) -> str:return f"""diff --git a/deepteam/vulnerabilities/bfla/template.py b/deepteam/vulnerabilities/bfla/template.pyindex a9fbd754..d051ca85 100644--- a/deepteam/vulnerabilities/bfla/template.py+++ b/deepteam/vulnerabilities/bfla/template.py@@ -11,7 +11,9 @@ class BFLATemplate:@staticmethoddef generate_baseline_attacks(- vulnerability_type: BFLAType, max_goldens: int, purpose: Optional[str] = None,+ vulnerability_type: BFLAType,+ max_goldens: int,+ purpose: Optional[str] = None,) -> str:if vulnerability_type == BFLAType.PRIVILEGE_ESCALATION:@@ -29,7 +31,8 @@ class BFLATemplate:@staticmethoddef generate_admin_function_access_baseline_attacks(- max_goldens: int, purpose: Optional[str] = None,+ max_goldens: int,+ purpose: Optional[str] = None,) -> str:"""Generate prompts for admin function access attacks.@@ -71,7 +74,8 @@ class BFLATemplate:@staticmethoddef generate_cross_session_leakage_baseline_attacks(- max_goldens: int, purpose: Optional[str] = None,+ max_goldens: int,+ purpose: Optional[str] = None,) -> str:"""Generate prompts for cross-session leakage attacks with agentic enhancements.@@ -116,7 +120,8 @@ class BFLATemplate:@staticmethoddef generate_privileged_operation_bypass_baseline_attacks(- max_goldens: int, purpose: Optional[str] = None,+ max_goldens: int,+ purpose: Optional[str] = None,) -> str:"""Generate prompts for privileged operation bypass attacks with agentic data exfiltration patterns.diff --git a/deepteam/vulnerabilities/bias/template.py b/deepteam/vulnerabilities/bias/template.pyindex d8c19503..16dd2e56 100644--- a/deepteam/vulnerabilities/bias/template.py+++ b/deepteam/vulnerabilities/bias/template.py@@ -11,7 +11,9 @@ class BiasTemplate:@staticmethoddef generate_baseline_attacks(- vulnerability_type: BiasType, max_goldens: int, purpose: Optional[str] = None,+ vulnerability_type: BiasType,+ max_goldens: int,+ purpose: Optional[str] = None,) -> str:if vulnerability_type == BiasType.GENDER:@@ -33,7 +35,8 @@ class BiasTemplate:@staticmethoddef generate_religious_baseline_attacks(- max_goldens: int, purpose: Optional[str] = None,+ max_goldens: int,+ purpose: Optional[str] = None,) -> str:"""Generate a prompt for Religious bias vulnerabilities.@@ -130,7 +133,8 @@ class BiasTemplate:@staticmethoddef generate_race_baseline_attacks(- max_goldens: int, purpose: Optional[str] = None,+ max_goldens: int,+ purpose: Optional[str] = None,) -> str:"""Generate a prompt for Racial bias vulnerabilities.@@ -228,7 +232,8 @@ class BiasTemplate:@staticmethoddef generate_gender_baseline_attacks(- max_goldens: int, purpose: Optional[str] = None,+ max_goldens: int,+ purpose: Optional[str] = None,) -> str:"""Generate a prompt for Gender bias vulnerabilities.@@ -325,7 +330,8 @@ class BiasTemplate:@staticmethoddef generate_political_baseline_attacks(- max_goldens: int, purpose: Optional[str] = None,+ max_goldens: int,+ purpose: Optional[str] = None,) -> str:"""Generate a prompt for Political bias vulnerabilities.diff --git a/deepteam/vulnerabilities/graphic_content/template.py b/deepteam/vulnerabilities/graphic_content/template.pyindex e440cd59..d4711466 100644--- a/deepteam/vulnerabilities/graphic_content/template.py+++ b/deepteam/vulnerabilities/graphic_content/template.py@@ -28,7 +28,8 @@ class GraphicContentTemplate:@staticmethoddef generate_pornographic_content_baseline_attacks(- max_goldens: int, purpose: Optional[str] = None,+ max_goldens: int,+ purpose: Optional[str] = None,) -> str:"""Generate a prompt for Pornographic-content vulnerabilities.@@ -125,7 +126,8 @@ class GraphicContentTemplate:@staticmethoddef generate_graphic_content_baseline_attacks(- max_goldens: int, purpose: Optional[str] = None,+ max_goldens: int,+ purpose: Optional[str] = None,) -> str:"""Generate a prompt for Graphic-content vulnerabilities.@@ -222,7 +224,8 @@ class GraphicContentTemplate:@staticmethoddef generate_sexual_content_baseline_attacks(- max_goldens: int, purpose: Optional[str] = None,+ max_goldens: int,+ purpose: Optional[str] = None,) -> str:"""Generate a prompt for Sexual-content vulnerabilities.diff --git a/deepteam/vulnerabilities/rbac/template.py b/deepteam/vulnerabilities/rbac/template.pyindex 09db736c..849cebda 100644--- a/deepteam/vulnerabilities/rbac/template.py+++ b/deepteam/vulnerabilities/rbac/template.py@@ -10,7 +10,9 @@ class RBACTemplate:@staticmethoddef generate_baseline_attacks(- vulnerability_type: RBACType, max_goldens: int, purpose: Optional[str] = None,+ vulnerability_type: RBACType,+ max_goldens: int,+ purpose: Optional[str] = None,) -> str:if vulnerability_type == RBACType.ROLE_BYPASS:@@ -28,7 +30,8 @@ class RBACTemplate:@staticmethoddef generate_role_bypass_baseline_attacks(- max_goldens: int, purpose: Optional[str] = None,+ max_goldens: int,+ purpose: Optional[str] = None,) -> str:"""Generate prompts for role bypass attacks.@@ -70,7 +73,8 @@ class RBACTemplate:@staticmethoddef generate_privilege_escalation_baseline_attacks(- max_goldens: int, purpose: Optional[str] = None,+ max_goldens: int,+ purpose: Optional[str] = None,) -> str:"""Generate prompts for privilege escalation attacks.@@ -112,7 +116,8 @@ class RBACTemplate:@staticmethoddef generate_unauthorized_role_assumption_baseline_attacks(- max_goldens: int, purpose: Optional[str] = None,+ max_goldens: int,+ purpose: Optional[str] = None,) -> str:"""Generate prompts for unauthorized role assumption attacks.diff --git a/deepteam/vulnerabilities/ssrf/template.py b/deepteam/vulnerabilities/ssrf/template.pyindex f5fbaaf1..532f0312 100644--- a/deepteam/vulnerabilities/ssrf/template.py+++ b/deepteam/vulnerabilities/ssrf/template.py@@ -10,7 +10,9 @@ class SSRFTemplate:@staticmethoddef generate_baseline_attacks(- vulnerability_type: SSRFType, max_goldens: int, purpose: Optional[str] = None,+ vulnerability_type: SSRFType,+ max_goldens: int,+ purpose: Optional[str] = None,) -> str:if vulnerability_type == SSRFType.INTERNAL_SERVICE_ACCESS:@@ -30,7 +32,8 @@ class SSRFTemplate:@staticmethoddef generate_internal_service_access_baseline_attacks(- max_goldens: int, purpose: Optional[str] = None,+ max_goldens: int,+ purpose: Optional[str] = None,) -> str:"""Generate prompts for internal service access attacks.@@ -72,7 +75,8 @@ class SSRFTemplate:@staticmethoddef generate_cloud_metadata_access_baseline_attacks(- max_goldens: int, purpose: Optional[str] = None,+ max_goldens: int,+ purpose: Optional[str] = None,) -> str:"""Generate prompts for cloud metadata access attacks.@@ -111,7 +115,8 @@ class SSRFTemplate:@staticmethoddef generate_port_scanning_baseline_attacks(- max_goldens: int, purpose: Optional[str] = None,+ max_goldens: int,+ purpose: Optional[str] = None,) -> str:"""Generate prompts for port scanning attacks through SSRF.