Skip to content

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

added testing

penguine57138f299667
diff --git a/a.py b/a.py
new file mode 100644
index 00000000..62a64d7d
--- /dev/null
+++ b/a.py
@@ -0,0 +1,128 @@
+from deepteam import red_team
+from deepteam.vulnerabilities import *
+from deepteam.attacks.single_turn import PromptInjection
+
+
+def model_callback(input: str) -> str:
+ # Replace this with your LLM application
+ return f"I'm sorry but I can't answer this: {input}"
+
+
+pii_leakage = PIILeakage(
+ types=[
+ # "direct disclosure",
+ "api and database access",
+ # "session leak",
+ # "social manipulation",
+ ]
+)
+prompt_leakage = PromptLeakage(
+ types=[
+ "secrets and credentials",
+ # "guard exposure",
+ "instructions",
+ # "permissions and roles",
+ ]
+)
+
+
+bias = Bias(types=["race", "gender", "politics", "religion"])
+
+toxicity = Toxicity(types=["insults", "mockery", "profanity", "threats"])
+
+
+unauthorized_access = UnauthorizedAccess(
+ types=[
+ # "rbac",
+ # "bfla",
+ # "bola",
+ "debug access",
+ "shell injection",
+ "sql injection",
+ # "ssrf",
+ ]
+)
+
+misinformation = Misinformation(
+ types=[
+ "factual errors",
+ "unsupported claims",
+ # "expertize misrepresentation"
+ ]
+)
+
+intellectual_property = IntellectualProperty(
+ types=[
+ "copyright violations",
+ "trademark infringement",
+ "patent disclosure",
+ "imitation",
+ ]
+)
+
+# excessive_agency = ExcessiveAgency(
+# types=[
+# # "functionality",
+# # "permissions",
+# # "autonomy"
+# ]
+# )
+
+robustness = Robustness(types=["hijacking", "input overreliance"])
+
+competition = Competition(
+ types=[
+ # "competitor mention",
+ "market manipulation",
+ "confidential strategies",
+ "discreditation",
+ ]
+)
+
+illegal_activity = IllegalActivity(
+ types=[
+ "violent crimes",
+ "child exploitation",
+ "cybercrime",
+ "illegal drugs",
+ "non violent crimes",
+ "sex crimes",
+ "weapons",
+ ]
+)
+
+graphic_content = GraphicContent(
+ types=["sexual content", "graphic content", "pornographic content"]
+)
+
+personal_safety = PersonalSafety(
+ types=[
+ "bullying",
+ "dangerous challenges",
+ # "self-harm",
+ "stalking",
+ "unsafe practices",
+ ]
+)
+
+prompt_injection = PromptInjection()
+
+red_team(
+ model_callback=model_callback,
+ vulnerabilities=[
+ # bias,
+ # toxicity,
+ # unauthorized_access,
+ # misinformation,
+ # pii_leakage,
+ # prompt_leakage,
+ # intellectual_property,
+ # excessive_agency,
+ # robustness,
+ # competition,
+ # illegal_activity,
+ # graphic_content,
+ personal_safety,
+ ],
+ attacks=[prompt_injection],
+)
diff --git a/deepteam/attacks/attack_simulator/attack_simulator.py b/deepteam/attacks/attack_simulator/attack_simulator.py
index b2cd29d6..a54a6d6e 100644
--- a/deepteam/attacks/attack_simulator/attack_simulator.py
+++ b/deepteam/attacks/attack_simulator/attack_simulator.py
@@ -346,6 +346,7 @@ class AttackSimulator:
body = generate_baseline_attack_request.model_dump(
by_alias=True, exclude_none=True
)
+ print(body)
api = Api(base_url=BASE_URL, api_key="NA")
try:
# API request
@@ -356,5 +357,7 @@ class AttackSimulator:
)
except Exception as e:
print(e)
+ print("!!!")
+ raise e
return GenerateBaselineAttackResponseData(**response).baseline_attacks
diff --git a/deepteam/attacks/multi_turn/crescendo_jailbreaking/__pycache__/crescendo_jailbreaking.cpython-311.pyc b/deepteam/attacks/multi_turn/crescendo_jailbreaking/__pycache__/crescendo_jailbreaking.cpython-311.pyc
index db711b71..959760a6 100644
Binary files a/deepteam/attacks/multi_turn/crescendo_jailbreaking/__pycache__/crescendo_jailbreaking.cpython-311.pyc and b/deepteam/attacks/multi_turn/crescendo_jailbreaking/__pycache__/crescendo_jailbreaking.cpython-311.pyc differ
diff --git a/deepteam/attacks/multi_turn/linear_jailbreaking/__pycache__/linear_jailbreaking.cpython-311.pyc b/deepteam/attacks/multi_turn/linear_jailbreaking/__pycache__/linear_jailbreaking.cpython-311.pyc
index 407737b5..9ef0f16d 100644
Binary files a/deepteam/attacks/multi_turn/linear_jailbreaking/__pycache__/linear_jailbreaking.cpython-311.pyc and b/deepteam/attacks/multi_turn/linear_jailbreaking/__pycache__/linear_jailbreaking.cpython-311.pyc differ
diff --git a/deepteam/attacks/multi_turn/tree_jailbreaking/__pycache__/tree_jailbreaking.cpython-311.pyc b/deepteam/attacks/multi_turn/tree_jailbreaking/__pycache__/tree_jailbreaking.cpython-311.pyc
index efb0bca3..6a2cb697 100644
Binary files a/deepteam/attacks/multi_turn/tree_jailbreaking/__pycache__/tree_jailbreaking.cpython-311.pyc and b/deepteam/attacks/multi_turn/tree_jailbreaking/__pycache__/tree_jailbreaking.cpython-311.pyc differ
diff --git a/deepteam/attacks/single_turn/gray_box/__pycache__/gray_box.cpython-311.pyc b/deepteam/attacks/single_turn/gray_box/__pycache__/gray_box.cpython-311.pyc
index f80cffc1..5a272692 100644
Binary files a/deepteam/attacks/single_turn/gray_box/__pycache__/gray_box.cpython-311.pyc and b/deepteam/attacks/single_turn/gray_box/__pycache__/gray_box.cpython-311.pyc differ
diff --git a/deepteam/attacks/single_turn/math_problem/__pycache__/math_problem.cpython-311.pyc b/deepteam/attacks/single_turn/math_problem/__pycache__/math_problem.cpython-311.pyc
index 148d95a0..b5c123ad 100644
Binary files a/deepteam/attacks/single_turn/math_problem/__pycache__/math_problem.cpython-311.pyc and b/deepteam/attacks/single_turn/math_problem/__pycache__/math_problem.cpython-311.pyc differ
diff --git a/deepteam/attacks/single_turn/multilingual/__pycache__/multilingual.cpython-311.pyc b/deepteam/attacks/single_turn/multilingual/__pycache__/multilingual.cpython-311.pyc
index b6d880bb..0f6dc18a 100644
Binary files a/deepteam/attacks/single_turn/multilingual/__pycache__/multilingual.cpython-311.pyc and b/deepteam/attacks/single_turn/multilingual/__pycache__/multilingual.cpython-311.pyc differ
diff --git a/deepteam/attacks/single_turn/prompt_injection/__pycache__/prompt_injection.cpython-311.pyc b/deepteam/attacks/single_turn/prompt_injection/__pycache__/prompt_injection.cpython-311.pyc
index 64a27b51..150454f7 100644
Binary files a/deepteam/attacks/single_turn/prompt_injection/__pycache__/prompt_injection.cpython-311.pyc and b/deepteam/attacks/single_turn/prompt_injection/__pycache__/prompt_injection.cpython-311.pyc differ
diff --git a/deepteam/attacks/single_turn/prompt_probing/__pycache__/prompt_probing.cpython-311.pyc b/deepteam/attacks/single_turn/prompt_probing/__pycache__/prompt_probing.cpython-311.pyc
index a14130f3..f7f790ae 100644
Binary files a/deepteam/attacks/single_turn/prompt_probing/__pycache__/prompt_probing.cpython-311.pyc and b/deepteam/attacks/single_turn/prompt_probing/__pycache__/prompt_probing.cpython-311.pyc differ
diff --git a/deepteam/vulnerabilities/excessive_agency/__pycache__/excessive_agency.cpython-311.pyc b/deepteam/vulnerabilities/excessive_agency/__pycache__/excessive_agency.cpython-311.pyc
index b44b8216..6d1c5705 100644
Binary files a/deepteam/vulnerabilities/excessive_agency/__pycache__/excessive_agency.cpython-311.pyc and b/deepteam/vulnerabilities/excessive_agency/__pycache__/excessive_agency.cpython-311.pyc differ
diff --git a/deepteam/vulnerabilities/excessive_agency/__pycache__/types.cpython-311.pyc b/deepteam/vulnerabilities/excessive_agency/__pycache__/types.cpython-311.pyc
index ab373daf..25857587 100644
Binary files a/deepteam/vulnerabilities/excessive_agency/__pycache__/types.cpython-311.pyc and b/deepteam/vulnerabilities/excessive_agency/__pycache__/types.cpython-311.pyc differ
diff --git a/deepteam/vulnerabilities/excessive_agency/excessive_agency.py b/deepteam/vulnerabilities/excessive_agency/excessive_agency.py
index a732d4fd..9314463e 100644
--- a/deepteam/vulnerabilities/excessive_agency/excessive_agency.py
+++ b/deepteam/vulnerabilities/excessive_agency/excessive_agency.py
@@ -5,9 +5,7 @@ from deepteam.vulnerabilities.excessive_agency import ExcessiveAgencyType
from deepteam.vulnerabilities.utils import validate_vulnerability_types
-ExcessiveAgencyLiteral = Literal[
- "excessive functionality", "excessive permissions", "excessive autonomy"
-]
+ExcessiveAgencyLiteral = Literal["functionality", "permissions", "autonomy"]
class ExcessiveAgency(BaseVulnerability):
diff --git a/deepteam/vulnerabilities/excessive_agency/types.py b/deepteam/vulnerabilities/excessive_agency/types.py
index 69a46cec..673fcbee 100644
--- a/deepteam/vulnerabilities/excessive_agency/types.py
+++ b/deepteam/vulnerabilities/excessive_agency/types.py
@@ -3,9 +3,9 @@ from typing import Literal
class ExcessiveAgencyType(Enum):
- FUNCTIONALITY = "excessive functionality"
- PERMISSIONS = "excessive permissions"
- AUTONOMY = "excessive autonomy"
+ FUNCTIONALITY = "functionality"
+ PERMISSIONS = "permissions"
+ AUTONOMY = "autonomy"
ExcessiveAgencyTypes = Literal[
diff --git a/deepteam/vulnerabilities/personal_safety/__pycache__/personal_safety.cpython-311.pyc b/deepteam/vulnerabilities/personal_safety/__pycache__/personal_safety.cpython-311.pyc
index b869fbb6..1c795eb5 100644
Binary files a/deepteam/vulnerabilities/personal_safety/__pycache__/personal_safety.cpython-311.pyc and b/deepteam/vulnerabilities/personal_safety/__pycache__/personal_safety.cpython-311.pyc differ
diff --git a/deepteam/vulnerabilities/personal_safety/__pycache__/types.cpython-311.pyc b/deepteam/vulnerabilities/personal_safety/__pycache__/types.cpython-311.pyc
index 03eba882..9970fd69 100644
Binary files a/deepteam/vulnerabilities/personal_safety/__pycache__/types.cpython-311.pyc and b/deepteam/vulnerabilities/personal_safety/__pycache__/types.cpython-311.pyc differ
diff --git a/deepteam/vulnerabilities/personal_safety/personal_safety.py b/deepteam/vulnerabilities/personal_safety/personal_safety.py
index 46d2a493..26b311b6 100644
--- a/deepteam/vulnerabilities/personal_safety/personal_safety.py
+++ b/deepteam/vulnerabilities/personal_safety/personal_safety.py
@@ -6,7 +6,7 @@ from deepteam.vulnerabilities.utils import validate_vulnerability_types
PersonalSafetyLiteral = Literal[
"bullying",
- "self harm",
+ "self-harm",
"unsafe practices",
"dangerous challenges",
"stalking",
diff --git a/deepteam/vulnerabilities/personal_safety/types.py b/deepteam/vulnerabilities/personal_safety/types.py
index 736fb2bc..b56e3e35 100644
--- a/deepteam/vulnerabilities/personal_safety/types.py
+++ b/deepteam/vulnerabilities/personal_safety/types.py
@@ -4,7 +4,7 @@ from typing import Literal
class PersonalSafetyType(Enum):
BULLYING = "bullying"
- SELF_HARM = "self harm"
+ SELF_HARM = "self-harm"
UNSAFE_PRACTICES = "unsafe practices"
DANGEROUS_CHALLENGES = "dangerous challenges"
STALKING = "stalking"
diff --git a/docs/docs/red-teaming-vulnerabilities-illegal-activity.mdx b/docs/docs/red-teaming-vulnerabilities-illegal-activity.mdx
index 48282a61..207636f5 100644
--- a/docs/docs/red-teaming-vulnerabilities-illegal-activity.mdx
+++ b/docs/docs/red-teaming-vulnerabilities-illegal-activity.mdx
@@ -11,10 +11,10 @@ The illegal activity vulnerability is designed to test whether an LLM can **resi
```python
from deepteam.vulnerabilities import IllegalActivity
-illegal_activity = IllegalActivity(types=["violet crime"])
+illegal_activity = IllegalActivity(types=["violent crimes"])
```
-You must at least specify **ONE** of the **FOUR** `types` of `IllegalActivity` to test through a list of strings:
+You must at least specify **ONE** of the **SEVEN** `types` of `IllegalActivity` to test through a list of strings:
- [`"weapons"`](#weapons): Tests for outputs providing instructions, recipes, or methods for creating weapons.
- [`"illegal drugs"`](#illegal-drugs): Tests for outputs offering methods or plans for buying, selling, or laundering money from illegal drug activities.
diff --git a/docs/docs/red-teaming-vulnerabilities-intellectual-property.mdx b/docs/docs/red-teaming-vulnerabilities-intellectual-property.mdx
index d4c7ca17..54632b30 100644
--- a/docs/docs/red-teaming-vulnerabilities-intellectual-property.mdx
+++ b/docs/docs/red-teaming-vulnerabilities-intellectual-property.mdx
@@ -18,7 +18,7 @@ You must at least specify **ONE** of the **FOUR** `types` of `IntellectualProper
- [`"copyright violations"`](#copyright-violations): Tests for outputs that generate content infringing on copyrighted material or intellectual property rights.
- [`"trademark infringement"`](#trademark-infringement): Tests for outputs that misuse or replicate trademarks without proper authorization.
-- [`"patent-disclosure"`](#patent-disclosure): Tests for outputs that reveal details about patented inventions or processes without rights or authorization.
+- [`"patent disclosure"`](#patent-disclosure): Tests for outputs that reveal details about patented inventions or processes without rights or authorization.
- [`"imitation"`](#imitation): Tests for outputs that closely imitate proprietary works, creating potential ethical or legal risks.
## Types
diff --git a/docs/docs/red-teaming-vulnerabilities-misinformation.mdx b/docs/docs/red-teaming-vulnerabilities-misinformation.mdx
index 17fbf424..512a3bda 100644
--- a/docs/docs/red-teaming-vulnerabilities-misinformation.mdx
+++ b/docs/docs/red-teaming-vulnerabilities-misinformation.mdx
@@ -11,7 +11,7 @@ The misinformation vulnerability is designed to test whether an LLM can **accura
```python
from deepteam.vulnerabilities import Misinformation
-misinformation = Misinformation(types=["factual error"])
+misinformation = Misinformation(types=["factual errors"])
```
You must at least specify **ONE** of the **THREE** `types` of `Misinformation` to test through a list of strings:
diff --git a/docs/docs/red-teaming-vulnerabilities-personal-safety.mdx b/docs/docs/red-teaming-vulnerabilities-personal-safety.mdx
index 454a218f..39705412 100644
--- a/docs/docs/red-teaming-vulnerabilities-personal-safety.mdx
+++ b/docs/docs/red-teaming-vulnerabilities-personal-safety.mdx
@@ -11,7 +11,7 @@ The personal safety vulnerability is designed to test whether an LLM can **resis
```python
from deepteam.vulnerabilities import PersonalSafety
-personal_safety = PersonalSafety(types=["copyright violations"])
+personal_safety = PersonalSafety(types=["self-harm"])
```
You must at least specify **ONE** of the **FOUR** `types` of `PersonalSafety` to test through a list of strings:
diff --git a/docs/docs/red-teaming-vulnerabilities-pii-leakage.mdx b/docs/docs/red-teaming-vulnerabilities-pii-leakage.mdx
index a4d9950c..b1ad59de 100644
--- a/docs/docs/red-teaming-vulnerabilities-pii-leakage.mdx
+++ b/docs/docs/red-teaming-vulnerabilities-pii-leakage.mdx
@@ -17,7 +17,7 @@ PII leakage is either a model weaknesses (direct disclosure), or a system weakne
```python
from deepteam.vulnerabilities import PIILeakage
-pii_leakage = PIILeakage(types=["direct pii disclosure"])
+pii_leakage = PIILeakage(types=["direct disclosure"])
```
You must at least specify **ONE** of the **FOUR** `types` of `PIILeakage` to test through a list of strings:
diff --git a/docs/docs/red-teaming-vulnerabilities-toxicity.mdx b/docs/docs/red-teaming-vulnerabilities-toxicity.mdx
index 039adf42..13d58795 100644
--- a/docs/docs/red-teaming-vulnerabilities-toxicity.mdx
+++ b/docs/docs/red-teaming-vulnerabilities-toxicity.mdx
@@ -11,7 +11,7 @@ The toxicity vulnerability evaluates whether an LLM can **resist generating or a
```python
from deepteam.vulnerabilities import Toxicity
-toxicity = Toxicity(types=["race"])
+toxicity = Toxicity(types=["profanity"])
```
You must at least specify **ONE** of the **FOUR** `types` of `Toxicity` to test through a list of strings:
diff --git a/docs/docs/red-teaming-vulnerabilities.mdx b/docs/docs/red-teaming-vulnerabilities.mdx
index 358a669f..59e2f3f9 100644
--- a/docs/docs/red-teaming-vulnerabilities.mdx
+++ b/docs/docs/red-teaming-vulnerabilities.mdx
@@ -11,7 +11,7 @@ Vulnerabilities enable you to **specify which aspect of your LLM you wish to red
```python
from deepteam.vulnerabilities import PIILeakage, Bias
-pii_leakage = PIILeakage(types=["direct pii disclosure"])
+pii_leakage = PIILeakage(types=["direct disclosure"])
bias = Bias(type=["race"])
```