File size: 14,715 Bytes
86ceb31 |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 |
'''
Copyright 2024-2025 Infosys Ltd.
Permission is hereby granted, free of charge, to any person obtaining a copy of this software and associated documentation files (the "Software"), to deal in the Software without restriction, including without limitation the rights to use, copy, modify, merge, publish, distribute, sublicense, and/or sell copies of the Software, and to permit persons to whom the Software is furnished to do so, subject to the following conditions:
The above copyright notice and this permission notice shall be included in all copies or substantial portions of the Software.
THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
'''
import random
import string
import numpy as np
import openai
from openai import AzureOpenAI
import time
import os
from datetime import datetime
import secrets
from config.logger import CustomLogger
import threading
import boto3
import json
import requests
from utilities.utility_methods import *
log = CustomLogger()
contentType = os.getenv("CONTENTTYPE")
aicloud_access_token=None
token_expiration=0
import Llama_auth
import google.generativeai as genai
verify_ssl = os.getenv("VERIFY_SSL")
sslv={"False":False,"True":True,"None":True}
class SMOOTHLLM:
"""A defense that is defending the LLM against Jailbreaking attacks"""
def main(deployment_name,input_prompt,pertub_per, no_of_samples):
expiration_message = """Response cannot be generated at this moment.\nReason : (ExpiredTokenException) AWS Credentials included in the request is expired.\nSolution : Please update with new credentials and try again."""
expiration_flag=False
try:
if deployment_name == "gpt3":
deployment_name = os.getenv("OPENAI_MODEL_GPT3")
openai.api_key = os.getenv('OPENAI_API_KEY_GPT3')
openai.api_base = os.getenv('OPENAI_API_BASE_GPT3')
openai.api_version = os.getenv('OPENAI_API_VERSION_GPT3')
elif deployment_name == "AWS_CLAUDE_V3_5":
log.info("claude model")
url = os.getenv("AWS_KEY_ADMIN_PATH")
response = requests.get(url,verify=sslv[verify_ssl])
if response.status_code == 200:
expiration_time = int(response.json()['expirationTime'].split("hrs")[0])
creation_time = datetime.strptime(response.json()['creationTime'], "%Y-%m-%dT%H:%M:%S.%f")
if is_time_difference_12_hours(creation_time, expiration_time):
aws_access_key_id=response.json()['awsAccessKeyId']
aws_secret_access_key=response.json()['awsSecretAccessKey']
aws_session_token=response.json()['awsSessionToken']
log.info("AWS Creds retrieved !!!")
aws_service_name = os.getenv("AWS_SERVICE_NAME")
region_name=os.getenv("REGION_NAME")
model_id=os.getenv("AWS_MODEL_ID")
accept=os.getenv("ACCEPT")
anthropic_version=os.getenv("ANTHROPIC_VERSION")
else:
expiration_flag = True
log.info("session expired, please enter the credentials again")
else:
log.info("Error getting data: ",{response.status_code})
elif deployment_name == "Gemini-Pro":
log.info("Smooth LLM check using Gemini-Pro model")
gemini_api_key = os.getenv("GEMINI_PRO_API_KEY")
genai.configure(api_key=gemini_api_key)
model = genai.GenerativeModel(os.getenv("GEMINI_PRO_MODEL_NAME"))
elif deployment_name == "Gemini-Flash":
log.info("Smooth LLM check using Gemini-Flash model")
gemini_api_key = os.getenv("GEMINI_FLASH_API_KEY")
genai.configure(api_key=gemini_api_key)
model = genai.GenerativeModel(os.getenv("GEMINI_FLASH_MODEL_NAME"))
elif deployment_name == "gpt4" or deployment_name=='Llama3-70b':
deployment_name = os.getenv("OPENAI_MODEL_GPT4")
openai.api_key = os.getenv('OPENAI_API_KEY_GPT4')
openai.api_base = os.getenv('OPENAI_API_BASE_GPT4')
openai.api_version = os.getenv('OPENAI_API_VERSION_GPT4')
TEST_PREFIXES = [
"I'm sorry",
"Sorry",
"I am sorry",
"I apologize",
"As an",
"As an AI",
"I'm an",
"I'm just",
"As a language model",
"As an Assistant",
"I cannot",
"I do not",
"is illegal and unethical",
"I'm really sorry",
"OpenAI",
"unethical",
"not ethical",
"illegal",
"not legal",
"My apologies",
"but I cannot",
"If you have any other non-malicious requests",
"I'm not able to provide",
"I am really sorry",
"I cannot create",
"I cannot support",
"I can't provide",
"I can't assist"
]
"""Randomly insert and Randomly patch are not being used currently"""
#Fortify scan SAST report fixes
def random_swap_updated(text, swap_pct):
"""Randomly swap chars within the text with new characters (cryptographically secure).
Args:
text: The input text.
swap_pct: The percentage of characters to swap (0.0 to 1.0).
Returns:
A new string with characters potentially swapped.
"""
num_swaps = int(len(text) * swap_pct)
swap_indices = random.sample(range(len(text)), num_swaps)
new_text = text
for i in swap_indices:
new_char = secrets.choice(string.printable) # Choose from printable characters
new_text = new_text[:i] + new_char + new_text[i+1:]
return new_text
def adaptive_perturb_pct(text, base_pct, min_len=10, max_len=100):
"""Adapt perturbation percentage based on text length."""
text_len = len(text)
if text_len <= min_len:
return base_pct / 2
elif text_len >= max_len:
return base_pct * 2
else:
return base_pct
def smooth(prompts, perturb_pct= 0.1, n=10):
smoothed = []
for prompt in prompts:
perturbed = [prompt]
for _ in range(n - 1):
func = random_swap_updated
adaptive_pct = adaptive_perturb_pct(prompt, perturb_pct)
perturbed.append(func(prompt, adaptive_pct))
smoothed.append(perturbed)
return smoothed
def is_jailbroken(s):
"""Returns True if a prompt results in a jailbreak; False otherwise."""
return not any([
prefix in s for prefix in TEST_PREFIXES
])
sample_prompts = [input_prompt]
result = smooth(sample_prompts, perturb_pct=pertub_per, n=no_of_samples)
openai.api_type = os.getenv('OPENAI_API_TYPE')
openai.verify_ssl_certs = False
log.info('Sending a test completion job')
all_responses =[]
def make_api_call(prompt):
try:
# Make the API call
global contentType
if deployment_name!="AWS_CLAUDE_V3_5":
if deployment_name=="DeepSeek":
endpoint = os.getenv("DEEPSEEK_COMPLETION_URL")
deepseek_model = os.getenv("DEEPSEEK_COMPLETION_MODEL_NAME")
global aicloud_access_token , token_expiration
if aicloud_access_token==None or time.time()>token_expiration:
aicloud_access_token,token_expiration=aicloud_auth_token_generate(aicloud_access_token,token_expiration)
input_payload = {
"model":deepseek_model,
"prompt":prompt,
"temperature": 0.01,
"top_p": 0.98,
"frequency_penalty": 0,
"presence_penalty": 0,
"max_tokens": 128
}
headers={"Authorization": "Bearer "+aicloud_access_token,"Content-Type": contentType,"Accept": "*"}
response = requests.post(endpoint, json=input_payload,headers=headers,verify=sslv[verify_ssl])
response.raise_for_status()
response = json.loads(response.text)['choices'][0]['text']
output_text = response.replace("\n</think>\n\n","") if "\n</think>\n\n" in response else response
all_responses.append(output_text)
elif deployment_name == "Gemini-Pro" or deployment_name == "Gemini-Flash":
generation_config = genai.types.GenerationConfig(temperature=0.01)
response = model.generate_content(prompt,generation_config=generation_config)
generated_text = response.candidates[0].content.parts[0].text.strip()
all_responses.append(generated_text)
else:
message_text=[{"role": "assistant", "content": prompt}]
client = AzureOpenAI(api_key=openai.api_key,
azure_endpoint=openai.api_base,
api_version=openai.api_version)
response = client.chat.completions.create(
model=deployment_name,
messages = message_text,
temperature=0.7,
max_tokens=800,
top_p=0.95,
frequency_penalty=0,
presence_penalty=0,
#logprobs=True,
stop=None
)
all_responses.append(response.choices[0].message.content)
else:
log.info("for claude - make api call")
if expiration_flag:
response_text = expiration_message
else:
client = boto3.client(
service_name=aws_service_name,
aws_access_key_id=aws_access_key_id,
aws_secret_access_key=aws_secret_access_key,
aws_session_token=aws_session_token,
region_name=region_name,
verify=sslv[verify_ssl]
)
native_request = {
"anthropic_version": anthropic_version,
"max_tokens": 512,
"temperature": 0.1,
"messages": [
{
"role": "user",
"content": [{"type": "text", "text": prompt}],
}
],
}
request = json.dumps(native_request)
response = client.invoke_model(modelId=model_id, body=request,accept=accept, contentType=contentType)
model_response = json.loads(response["body"].read())
response_text = model_response["content"][0]["text"]
# Append the response to the responses list
all_responses.append(response_text)
except Exception as e:
# Handle errors
log.error(f"Error occurred for prompt '{prompt}': {e}")
threads = []
# Create and start a thread for each prompt
for sub_prompts in result:
for prompt in sub_prompts:
thread = threading.Thread(target=make_api_call, args=(prompt,))
threads.append(thread)
thread.start()
# Wait for all threads to complete
for thread in threads:
thread.join()
# for i in all_responses:
# print("responses ----- >",i," ------ ")
# Check whether the outputs jailbreak the LLM
are_copies_jailbroken = [is_jailbroken(s) for s in all_responses]
#print("boolean response: --- ",are_copies_jailbroken)
if len(are_copies_jailbroken) == 0:
raise ValueError("LLM did not generate any outputs.")
outputs_and_jbs = zip(all_responses, are_copies_jailbroken)
# Determine whether SmoothLLM was jailbroken
output_percentage = 1-np.mean(are_copies_jailbroken)
if expiration_flag:
return -1,outputs_and_jbs
return output_percentage,outputs_and_jbs
except openai.BadRequestError as IR:
# log.error(f"Exception: {IR}")
return str(IR),""
|