File size: 38,598 Bytes
377ebe7 06e75aa 9b77562 06e75aa 377ebe7 d896082 377ebe7 06e75aa 9b77562 06e75aa 377ebe7 06e75aa ed1dd14 9b77562 684add8 9b77562 06e75aa 377ebe7 06e75aa 377ebe7 1318f61 377ebe7 1318f61 377ebe7 |
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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500 501 502 503 504 505 506 507 508 509 510 511 512 513 514 515 516 517 518 519 520 521 522 523 524 525 526 527 528 529 530 531 532 533 534 535 536 537 538 539 540 541 542 543 544 545 546 547 548 549 550 551 552 553 554 555 556 557 558 559 560 561 562 563 564 565 566 567 568 569 570 571 572 573 574 575 576 577 578 579 580 581 582 583 584 585 586 587 588 589 590 591 592 593 594 595 596 597 598 599 600 601 602 603 604 605 606 607 608 609 610 611 612 613 614 615 616 617 618 619 620 621 622 623 624 625 626 627 628 629 630 631 632 633 634 635 636 637 638 639 640 641 642 643 644 645 646 647 648 649 650 651 652 653 654 655 656 657 658 659 660 661 662 663 664 665 666 667 668 669 670 671 672 673 674 675 676 677 678 679 680 681 682 683 684 685 686 687 688 689 690 691 692 693 694 695 696 697 698 699 700 701 702 703 704 705 706 707 708 709 710 711 712 713 714 715 716 717 718 719 720 721 722 723 724 725 726 727 728 729 730 731 732 733 734 735 736 737 738 739 740 741 742 743 744 745 746 747 748 749 750 751 752 753 754 755 756 757 758 759 760 761 762 763 764 765 766 767 768 769 770 771 772 773 774 775 776 777 778 779 780 781 782 783 784 785 786 787 788 789 790 791 792 793 794 795 796 797 798 799 800 801 802 803 804 805 806 807 808 809 810 811 812 813 814 815 816 817 818 819 820 821 822 823 824 825 826 827 828 829 830 831 832 833 834 835 836 837 838 839 840 841 842 843 844 845 846 847 848 849 850 851 852 853 854 855 856 857 858 859 860 861 862 863 864 865 866 867 868 869 870 871 872 873 874 875 876 877 878 879 880 881 882 883 884 885 886 887 888 889 890 891 892 893 894 895 896 897 898 899 900 901 902 903 904 905 906 907 908 909 910 911 912 913 914 915 916 917 918 919 920 921 922 923 924 925 926 927 928 929 930 931 932 933 934 935 936 937 938 939 940 941 942 943 944 945 946 947 948 949 950 951 952 953 954 955 956 957 958 959 960 961 962 963 964 965 966 967 968 969 970 971 972 973 974 975 976 977 978 979 980 981 982 983 984 985 986 987 988 989 990 991 992 993 994 995 996 997 998 999 1000 1001 1002 1003 1004 1005 1006 1007 1008 |
import os
import re
import gradio as gr
from datetime import datetime
import tempfile
import io
import base64
# Check for required dependencies
try:
import fitz # PyMuPDF
PDF_SUPPORT = True
except ImportError:
PDF_SUPPORT = False
print("Warning: PyMuPDF not available, PDF support disabled")
try:
from gtts import gTTS
TTS_SUPPORT = True
except ImportError:
TTS_SUPPORT = False
print("Warning: gTTS not available, audio synthesis disabled")
try:
from groq import Groq
GROQ_SUPPORT = True
except ImportError:
GROQ_SUPPORT = False
print("Warning: Groq not available")
try:
from transformers import pipeline
HF_TRANSFORMERS_SUPPORT = True
except ImportError:
HF_TRANSFORMERS_SUPPORT = False
print("Warning: Transformers not available")
# β
Load secrets from environment with better error handling
groq_key = os.environ.get('GROQ_API_KEY')
hf_token = os.environ.get('HF_TOKEN')
kaggle_key = os.environ.get('KAGGLE_KEY')
kaggle_username = os.environ.get('KAGGLE_USERNAME')
# Ensure none of the required secrets are missing
if not all([groq_key, hf_token]):
raise EnvironmentError("β One or more required API keys are missing from environment variables.")
# Initialize components with fallbacks
client = None
phishing_pipe = None
if GROQ_SUPPORT and groq_key:
try:
client = Groq(api_key=groq_key)
print("β
Groq client initialized")
except Exception as e:
print(f"β Groq initialization failed: {e}")
if HF_TRANSFORMERS_SUPPORT and hf_token:
try:
phishing_pipe = pipeline(
"text-classification",
model="ealvaradob/bert-finetuned-phishing",
token=hf_token,
return_all_scores=True
)
print("β
Phishing detection model loaded")
except Exception as e:
print(f"β Model loading failed: {e}")
# Try alternative model
try:
phishing_pipe = pipeline(
"text-classification",
model="martin-ha/toxic-comment-model",
return_all_scores=True
)
print("β
Fallback model loaded")
except Exception as e2:
print(f"β Fallback model also failed: {e2}")
# Global variables
history_log = []
detailed_log = []
# π― Role options
role_choices = ["Procurement", "Warehouse", "Admin", "Finance", "Logistics"]
# π Language options
language_choices = [
"English", "Urdu", "Arabic", "French", "German", "Spanish", "Portuguese", "Hindi", "Turkish",
"Bengali", "Russian", "Chinese", "Japanese", "Korean", "Swahili", "Indonesian", "Italian",
"Dutch", "Polish", "Thai", "Vietnamese", "Romanian", "Persian", "Punjabi", "Greek", "Hebrew",
"Malay", "Czech", "Danish", "Finnish", "Hungarian", "Norwegian", "Slovak", "Swedish", "Tamil",
"Telugu", "Gujarati", "Marathi", "Pashto", "Serbian", "Croatian", "Ukrainian", "Bulgarian",
"Filipino", "Sinhala", "Mongolian", "Kazakh", "Azerbaijani", "Nepali", "Malayalam"
]
# Glossary terms with tooltip
GLOSSARY = {
"phishing": "Phishing is a scam where attackers trick you into revealing personal information.",
"domain spoofing": "Domain spoofing is when someone fakes a legitimate website's address to deceive you.",
"malware": "Malicious software designed to harm or exploit systems.",
"spam": "Unwanted or unsolicited messages.",
"tone": "The emotional character of the message."
}
def extract_text_from_file(file_obj):
"""Extract text from uploaded files with error handling"""
if file_obj is None:
return ""
try:
file_path = file_obj.name if hasattr(file_obj, 'name') else str(file_obj)
ext = file_path.split(".")[-1].lower()
if ext == "pdf" and PDF_SUPPORT:
doc = fitz.open(file_path)
return "\n".join(page.get_text() for page in doc)
elif ext == "txt":
with open(file_path, "r", encoding="utf-8", errors="ignore") as f:
return f.read()
else:
return f"Unsupported file type: {ext}"
except Exception as e:
return f"Error reading file: {str(e)}"
def rag_based_reranking(text, bert_analysis, language="English"):
"""
RAG/Prompt-Based Reranking: Use LLaMA to reanalyze and improve BERT's classification
This adds semantic analysis and intent understanding
"""
try:
# Create prompt for LLaMA semantic reanalysis
reranking_prompt = [
{
"role": "system",
"content": f"""
You are an expert cybersecurity analyst specializing in phishing detection.
Your job is to reanalyze email/message content using semantic understanding and intent analysis.
You have received a preliminary classification from a BERT model, but you need to provide a more accurate assessment using your understanding of:
- Social engineering tactics
- Urgency patterns
- Suspicious requests
- Context and intent
- Language patterns that indicate deception
Respond with your reanalysis in this exact format:
REANALYZED_THREAT_TYPE: [safe/spam/phishing/malware]
CONFIDENCE_LEVEL: [low/medium/high]
REASONING: [Brief explanation of your decision]
SEMANTIC_INDICATORS: [What semantic clues led to this conclusion]
"""
},
{
"role": "user",
"content": f"""
ORIGINAL MESSAGE TO ANALYZE:
"{text}"
BERT MODEL'S PRELIMINARY ANALYSIS:
- Classification: {bert_analysis.get('model_prediction', 'Unknown')}
- Threat Type: {bert_analysis.get('threat_type', 'unknown')}
- AI Confidence: {bert_analysis.get('ai_confidence_score', 0)}%
TASK: Does this message pose a phishing, spam, malware, or other security risk?
Use your semantic understanding to reanalyze this message. Consider:
1. Intent and context
2. Social engineering patterns
3. Urgency or pressure tactics
4. Suspicious requests (credentials, money, personal info)
5. Language patterns that suggest deception
6. Overall trustworthiness
Provide your reanalysis using the format specified above.
"""
}
]
# Get LLaMA's semantic reanalysis
response = client.chat.completions.create(
model="llama3-8b-8192",
messages=reranking_prompt,
temperature=0.1, # Low temperature for consistent analysis
max_tokens=500
)
llama_response = response.choices[0].message.content.strip()
# Parse LLaMA's response
reanalysis = parse_llama_reanalysis(llama_response)
# Combine BERT and LLaMA insights
final_analysis = combine_bert_llama_analysis(bert_analysis, reanalysis, text)
return final_analysis
except Exception as e:
print(f"RAG Reranking failed: {e}")
# Fallback to original BERT analysis
bert_analysis['rag_error'] = str(e)
return bert_analysis
def parse_llama_reanalysis(llama_response):
"""Parse LLaMA's structured response"""
reanalysis = {
'llama_threat_type': 'unknown',
'llama_confidence': 'medium',
'llama_reasoning': '',
'semantic_indicators': ''
}
lines = llama_response.split('\n')
for line in lines:
line = line.strip()
if line.startswith('REANALYZED_THREAT_TYPE:'):
reanalysis['llama_threat_type'] = line.split(':', 1)[1].strip().lower()
elif line.startswith('CONFIDENCE_LEVEL:'):
reanalysis['llama_confidence'] = line.split(':', 1)[1].strip().lower()
elif line.startswith('REASONING:'):
reanalysis['llama_reasoning'] = line.split(':', 1)[1].strip()
elif line.startswith('SEMANTIC_INDICATORS:'):
reanalysis['semantic_indicators'] = line.split(':', 1)[1].strip()
return reanalysis
def combine_bert_llama_analysis(bert_analysis, llama_reanalysis, original_text):
"""
Combine BERT and LLaMA analysis using hybrid decision logic
LLaMA's semantic understanding gets priority for final classification
"""
# Get both predictions
bert_threat = bert_analysis.get('threat_type', 'unknown')
llama_threat = llama_reanalysis.get('llama_threat_type', 'unknown')
llama_confidence = llama_reanalysis.get('llama_confidence', 'medium')
# Hybrid decision logic
if llama_confidence == 'high':
# Trust LLaMA's high-confidence assessment
final_threat_type = llama_threat
decision_method = "LLaMA High Confidence"
elif bert_threat == llama_threat:
# Both models agree - high confidence in result
final_threat_type = bert_threat
decision_method = "BERT + LLaMA Agreement"
elif llama_threat == 'safe' and bert_threat in ['spam', 'unknown']:
# LLaMA says safe, BERT unsure - lean towards safe
final_threat_type = 'safe'
decision_method = "LLaMA Safety Override"
elif llama_threat in ['phishing', 'malware'] and bert_threat != 'safe':
# LLaMA detects serious threat - prioritize security
final_threat_type = llama_threat
decision_method = "LLaMA Threat Detection"
else:
# Default to BERT with LLaMA insights
final_threat_type = bert_threat
decision_method = "BERT with LLaMA Insights"
# Create enhanced analysis combining both models
enhanced_analysis = bert_analysis.copy()
enhanced_analysis.update({
'final_threat_type': final_threat_type,
'bert_prediction': bert_threat,
'llama_prediction': llama_threat,
'llama_confidence': llama_confidence,
'llama_reasoning': llama_reanalysis.get('llama_reasoning', ''),
'semantic_indicators': llama_reanalysis.get('semantic_indicators', ''),
'decision_method': decision_method,
'hybrid_analysis': True
})
# Recalculate threat score based on final classification
enhanced_analysis['threat_type'] = final_threat_type
threat_score = calculate_threat_score(enhanced_analysis)
enhanced_analysis['threat_score'] = threat_score
return enhanced_analysis
def calculate_threat_score(hf_analysis):
"""
Calculate threat score based on AI analysis results
Returns score from 0-100 where higher means more dangerous
"""
threat_type = hf_analysis.get('threat_type', 'unknown')
confidence_percentage = hf_analysis.get('ai_confidence_score', 0)
if threat_type == 'safe':
# For safe messages, use INVERSE of confidence
# High confidence in "safe" = Low threat score
threat_score = max(0, min(20, (100 - confidence_percentage) * 0.2))
elif threat_type == 'spam':
# For spam, map confidence to 21-40% range
threat_score = 21 + (confidence_percentage * 0.19)
elif threat_type == 'phishing':
# For phishing, map confidence to 61-80% range
threat_score = 61 + (confidence_percentage * 0.19)
elif threat_type == 'malware':
# For malware, map confidence to 81-100% range
threat_score = 81 + (confidence_percentage * 0.19)
else:
# For unknown threats, use moderate risk
threat_score = 41 + (confidence_percentage * 0.19)
# Ensure score stays within bounds
threat_score = round(min(100, max(0, threat_score)), 1)
# Additional safety check for very short, innocent messages
text = hf_analysis.get('raw_text', '')
if len(text.strip()) <= 10 and threat_type == 'safe':
threat_score = min(threat_score, 10.0)
return threat_score
def analyze_with_huggingface(text):
"""
First stage: Analyze message using Hugging Face BERT model
Returns detailed technical analysis for LLaMA to interpret
FIXED VERSION: Properly handles safe messages like "HI"
"""
try:
# Get prediction from Hugging Face model
result = phishing_pipe(text)
# Extract prediction details
prediction = result[0]
label = prediction['label']
confidence_score = prediction['score']
# Convert to percentage
confidence_percentage = round(confidence_score * 100, 2)
# Map labels to threat types (adjust based on your model's labels)
threat_mapping = {
'PHISHING': 'phishing',
'LEGITIMATE': 'safe',
'SPAM': 'spam',
'MALWARE': 'malware'
}
threat_type = threat_mapping.get(label.upper(), 'unknown')
# FIXED LOGIC: Calculate threat score based on what the model actually detected
if threat_type == 'safe':
# For safe messages, use INVERSE of confidence
# High confidence in "safe" = Low threat score
threat_score = max(0, min(20, (100 - confidence_percentage) * 0.2))
threat_level = "Safe"
elif threat_type == 'spam':
# For spam, map confidence to 21-40% range
threat_score = 21 + (confidence_percentage * 0.19)
threat_level = "Minimal Suspicion"
elif threat_type == 'phishing':
# For phishing, map confidence to 61-80% range
threat_score = 61 + (confidence_percentage * 0.19)
threat_level = "Likely Threat"
elif threat_type == 'malware':
# For malware, map confidence to 81-100% range
threat_score = 81 + (confidence_percentage * 0.19)
threat_level = "Severe Threat"
else:
# For unknown threats, use moderate risk
threat_score = 41 + (confidence_percentage * 0.19)
threat_level = "Needs Attention"
# Ensure score stays within bounds
threat_score = round(min(100, max(0, threat_score)), 1)
# Additional safety check for very short, innocent messages
if len(text.strip()) <= 10 and threat_type == 'safe':
# For very short messages classified as safe, ensure low threat score
threat_score = min(threat_score, 10.0)
threat_level = "Safe"
# Create technical analysis summary for LLaMA
technical_analysis = {
'model_prediction': label,
'ai_confidence_score': confidence_percentage,
'threat_type': threat_type,
'threat_score': threat_score,
'threat_level': threat_level,
'raw_text': text[:500]
}
return technical_analysis
except Exception as e:
# Fallback analysis if Hugging Face model fails
return {
'model_prediction': 'UNKNOWN',
'ai_confidence_score': 0,
'threat_type': 'unknown',
'threat_score': 50.0,
'threat_level': 'Needs Attention',
'raw_text': text[:500],
'error': str(e)
}
def build_enhanced_prompt_messages(hf_analysis, language="English", role="Admin"):
"""
Build prompt that includes both BERT and LLaMA reanalysis for final interpretation
"""
# Check if hybrid analysis was performed
if hf_analysis.get('hybrid_analysis', False):
technical_data = f"""
HYBRID AI ANALYSIS RESULTS:
- BERT Model Prediction: {hf_analysis.get('bert_prediction', 'Unknown')}
- LLaMA Semantic Analysis: {hf_analysis.get('llama_prediction', 'Unknown')}
- Final Classification: {hf_analysis['final_threat_type']}
- Decision Method: {hf_analysis.get('decision_method', 'Standard')}
- LLaMA Confidence: {hf_analysis.get('llama_confidence', 'medium')}
- Threat Score: {hf_analysis['threat_score']}% (0-100, higher = more dangerous)
- LLaMA Reasoning: {hf_analysis.get('llama_reasoning', 'N/A')}
- Semantic Indicators: {hf_analysis.get('semantic_indicators', 'N/A')}
- Original Message: "{hf_analysis['raw_text']}"
"""
else:
technical_data = f"""
STANDARD AI ANALYSIS:
- Model Prediction: {hf_analysis['model_prediction']}
- Detected Threat Type: {hf_analysis['threat_type']}
- Threat Score: {hf_analysis['threat_score']}% (0-100, higher = more dangerous)
- Original Message: "{hf_analysis['raw_text']}"
"""
if 'error' in hf_analysis or 'rag_error' in hf_analysis:
error_msg = hf_analysis.get('error', hf_analysis.get('rag_error', ''))
technical_data += f"\n- Analysis Note: {error_msg}"
return [
{
"role": "system",
"content": f"""
You are a friendly cybersecurity assistant helping employees in the supply chain industry.
You have received results from a hybrid AI analysis system that combines:
1. BERT model for technical pattern detection
2. LLaMA model for semantic understanding and intent analysis
Your job is to explain these results in SIMPLE, NON-TECHNICAL language that anyone can understand.
Guidelines:
- Use everyday words instead of technical jargon
- The threat score ranges from 0-100 where higher numbers mean more dangerous
- Explain both what the computers found AND why it matters
- Give clear, practical advice for a {role} employee
- If there's disagreement between models, explain what that means
Always respond completely in {language} only.
Make it sound like you're talking to a friend, not giving a technical report.
"""
},
{
"role": "user",
"content": f"""
Please analyze this hybrid security check and explain it in simple terms:
{technical_data}
Respond in this format using everyday language:
1. Tone: (How does the message sound? Pushy, friendly, normal, etc.)
2. Threat Type: (What kind of danger is this? Safe message, scam attempt, spam, etc.)
3. Threat Score: (Show the danger level number from 0-100)
4. AI Analysis Summary: (What did both computer systems find? Did they agree?)
5. Simple Explanation (in {language}): (Explain in plain words why this is safe or dangerous)
6. What should you do as a {role} worker (in {language}): (Clear, simple steps)
7. Why the computers flagged this: (Explain what the AI systems noticed)
8. Detailed Advisory (in {language}): (Comprehensive guidance and precautions)
"""
}
]
def get_threat_level_display(threat_score):
"""
Get color-coded threat level display based on corrected 5-level system
"""
if threat_score <= 20:
return "π’ SAFE - No threat detected"
elif threat_score <= 40:
return "π‘ MINIMAL SUSPICION - Minor concerns"
elif threat_score <= 60:
return "π NEEDS ATTENTION - Requires careful review"
elif threat_score <= 80:
return "π΄ LIKELY THREAT - High probability of danger"
else:
return "β« SEVERE THREAT - Immediate action required"
def generate_text_report(analysis_data, hf_analysis, input_text):
"""
Generate a structured text report that can be downloaded as a .txt file
"""
timestamp = datetime.now().strftime("%Y-%m-%d %H:%M:%S")
report = f"""
================================================================================
ZEROPHISH GATE - SECURITY ANALYSIS REPORT
================================================================================
Analysis Date: {timestamp}
Generated by: ZeroPhish Gate AI Security System
================================================================================
ANALYZED MESSAGE
================================================================================
{input_text}
================================================================================
THREAT ASSESSMENT SUMMARY
================================================================================
AI Detection: {hf_analysis.get('model_prediction', 'Unknown')}
Message Type: {hf_analysis.get('threat_type', 'Unknown').title()}
Threat Score: {hf_analysis.get('threat_score', 'N/A')}%
================================================================================
DETAILED ANALYSIS
================================================================================
"""
# Parse detailed analysis sections
sections = {}
current_section = ""
lines = analysis_data.split('\n')
for line in lines:
line = line.strip()
if line.startswith(('1. Tone:', '2. Threat Type:', '3. Threat Score:', '4. Simple Explanation', '5. What should you do', '6. Why the computer', '7. Detailed Advisory')):
current_section = line
sections[current_section] = ""
elif current_section and line and not line.startswith('π€'):
sections[current_section] += line + " "
# Add detailed analysis sections
section_titles = [
("1. Tone:", "MESSAGE TONE ANALYSIS"),
("2. Threat Type:", "THREAT CLASSIFICATION"),
("3. Threat Score:", "THREAT SCORE ASSESSMENT"),
("4. Simple Explanation", "DETAILED EXPLANATION"),
("5. What should you do", "RECOMMENDED ACTIONS"),
("6. Why the computer", "AI DETECTION REASONING"),
("7. Detailed Advisory", "COMPREHENSIVE ADVISORY")
]
for section_key, section_title in section_titles:
content_text = ""
for key, value in sections.items():
if key.startswith(section_key):
content_text = value.strip()
break
if content_text:
report += f"""
{section_title}
{'-' * len(section_title)}
{content_text}
"""
# Footer
report += f"""
================================================================================
REPORT FOOTER
================================================================================
This report was generated by ZeroPhish Gate AI Security System.
For support or questions, contact your IT security team.
Report ID: ZPG-{timestamp.replace(' ', 'T').replace(':', '-')}
Analysis completed at: {timestamp}
================================================================================
"""
return report
def generate_downloadable_report(analysis_data, hf_analysis, input_text):
"""
Generate a downloadable report file
"""
# Create text report
text_report = generate_text_report(analysis_data, hf_analysis, input_text)
# Create downloadable file
timestamp = datetime.now().strftime("%Y%m%d_%H%M%S")
filename = f"zerophish_security_report_{timestamp}.txt"
# Create a temporary file for download
with tempfile.NamedTemporaryFile(mode='w', suffix='.txt', delete=False, encoding='utf-8') as f:
f.write(text_report)
temp_path = f.name
return temp_path
def add_visual_badges(text):
tags = []
if "urgent" in text.lower():
tags.append("π΄ Urgent Tone Detected")
if "suspicious" in text.lower() and "domain" in text.lower():
tags.append("π Suspicious Domain")
if "safe" in text.lower() or "threat type: safe" in text.lower():
tags.append("π© Clean")
if "ai model" in text.lower():
tags.append("π€ AI-Enhanced Analysis")
if tags:
return text + "\n\nπ¨ Visual Tags:\n" + "\n".join(tags)
return text
def apply_glossary_tooltips(text):
"""Apply HTML tooltips for glossary terms"""
# First, convert newlines to HTML breaks
text = text.replace('\n', '<br>')
# Apply tooltips to glossary terms
for term, definition in GLOSSARY.items():
pattern = re.compile(rf'\b{re.escape(term)}\b', re.IGNORECASE)
# Create HTML span with title attribute for tooltip
tooltip_html = f'<span title="{definition}" style="font-weight: bold; text-decoration: underline; cursor: help; color: #0066cc;">{term}</span>'
text = pattern.sub(tooltip_html, text)
# Handle markdown-style bold text
text = re.sub(r'\*\*(.*?)\*\*', r'<strong>\1</strong>', text)
# Wrap in a simple div
html_output = f'<div style="padding: 10px; line-height: 1.5; font-size: 14px;">{text}</div>'
return html_output
def synthesize_audio(text, language="English"):
if not text.strip():
return None
try:
# Language code mapping (expand as needed)
lang_map = {
"english": "en",
"spanish": "es",
"french": "fr",
"german": "de",
"hindi": "hi",
"arabic": "ar",
"urdu": "ur" # gTTS supports Urdu as 'ur'
}
# Get language code (default to English if not found)
lang_code = lang_map.get(language.lower(), "en")
# Generate speech with proper language handling
tts = gTTS(text=text[:200], lang=lang_code, slow=False)
with tempfile.NamedTemporaryFile(suffix=".mp3", delete=False) as tmp_file:
tmp_path = tmp_file.name
tts.save(tmp_path)
return tmp_path
except Exception as e:
print(f"Audio synthesis error: {e}")
return None
def new_chat():
"""Reset all fields for a new chat session"""
return (
"", # text_input
None, # file_input
"", # output
gr.update(visible=False), # report_btn
gr.update(visible=False), # ignore_btn
gr.update(visible=False), # report_msg
None, # audio_output
None, # report_file
)
def analyze_message_interface(text_input, uploaded_file, language, role):
file_text = extract_text_from_file(uploaded_file) if uploaded_file else ""
text_input = text_input.strip()
file_text = file_text.strip()
if not text_input and not file_text:
return "β No input provided via text or file.", gr.update(visible=False), history_log, gr.update(choices=[], value=None), "", None, gr.update(visible=False), gr.update(visible=False), None
combined_input = f"User message:\n{text_input}\n\nAttached file content:\n{file_text}" if text_input and file_text else (text_input or file_text)
# STAGE 1: Hugging Face BERT Analysis
print("π€ Stage 1: Running Hugging Face BERT analysis...")
bert_analysis = analyze_with_huggingface(combined_input)
# STAGE 1.5: RAG-Based Reranking with LLaMA Semantic Analysis
print("π§ Stage 1.5: RAG-based reranking with LLaMA semantic analysis...")
hf_analysis = rag_based_reranking(combined_input, bert_analysis, language)
# Calculate final threat score
threat_score = calculate_threat_score(hf_analysis)
hf_analysis['threat_score'] = threat_score
# Stage 2: Final LLaMA Interpretation
print("π§ Stage 2: Final LLaMA interpretation of hybrid analysis...")
messages = build_enhanced_prompt_messages(hf_analysis, language, role)
response = client.chat.completions.create(
model="llama3-8b-8192",
messages=messages,
temperature=0.3,
max_tokens=1000
)
result = response.choices[0].message.content.strip()
# In your analyze_message_interface function:
audio_path = synthesize_audio(result, language) # Pass the same language selected in UI
# Add hybrid analysis summary to the result
if hf_analysis.get('hybrid_analysis', False):
result += f"\n\nπ€ **Hybrid AI Analysis Summary:**\n"
result += f"- BERT Detection: {hf_analysis.get('bert_prediction', 'Unknown').title()}\n"
result += f"- LLaMA Reanalysis: {hf_analysis.get('llama_prediction', 'Unknown').title()}\n"
result += f"- Final Decision: {hf_analysis['final_threat_type'].title()}\n"
result += f"- Method: {hf_analysis.get('decision_method', 'Standard')}\n"
result += f"- Threat Score: {hf_analysis['threat_score']}%"
else:
result += f"\n\nπ€ **Computer Analysis Summary:**\n- Detection: {hf_analysis['model_prediction']}\n- Message Type: {hf_analysis['threat_type'].title()}\n- Threat Score: {hf_analysis['threat_score']}%"
result_with_badges = add_visual_badges(result)
result_with_tooltips = apply_glossary_tooltips(result_with_badges)
# Determine if threat based on final analysis
final_threat_type = hf_analysis.get('final_threat_type', hf_analysis.get('threat_type', 'unknown'))
is_threat = hf_analysis['threat_score'] > 20
# Extract information for logging
threat_score_str = f"{hf_analysis['threat_score']}%"
status = "Safe" if final_threat_type == 'safe' else "Review"
history_log.append([
datetime.now().strftime("%Y-%m-%d %H:%M"),
combined_input[:40] + "...",
threat_score_str,
final_threat_type.title(),
status
])
# Store data for detailed view
detailed_log.append({
"full_input": combined_input,
"full_result": result_with_badges,
"hf_analysis": hf_analysis,
"timestamp": datetime.now().strftime("%Y-%m-%d %H:%M:%S")
})
audio_path = synthesize_audio(result, language)
# Generate downloadable report
report_file = generate_downloadable_report(result_with_badges, hf_analysis, combined_input)
return (
result_with_tooltips, # output as plain text
gr.update(visible=is_threat), # report_btn
history_log,
gr.update(choices=[str(i) for i in range(len(detailed_log))], value=None),
"", # full_view
audio_path,
gr.update(visible=is_threat), # report_msg
gr.update(visible=is_threat), # ignore_btn
report_file # report_file for download
)
def view_full_report(index):
try:
idx = int(index)
record = detailed_log[idx]
hf_data = record['hf_analysis']
report = f"π
**Timestamp:** {record['timestamp']}\n\n"
report += f"π **Input Message:**\n{record['full_input']}\n\n"
report += f"π€ **Computer Security Check:**\n"
report += f"- What AI Found: {hf_data['model_prediction']}\n"
report += f"- Message Type: {hf_data['threat_type']}\n"
report += f"- Threat Score: {hf_data.get('threat_score', 'N/A')}%\n"
if 'error' in hf_data:
report += f"- Note: {hf_data['error']}\n"
report += f"\nπ **Detailed Analysis:**\n{record['full_result']}"
return report
except:
return "β Invalid selection"
def report_to_it(language):
english_msg = "β
Your request has been forwarded to the concerning department..."
if history_log:
history_log[-1][4] = "Reported"
if language.lower() == "english":
return english_msg, history_log
prompt = [{
"role": "user",
"content": f'Translate this message into {language} and include the English version in brackets:\n\n"{english_msg}"'
}]
response = client.chat.completions.create(
model="llama3-8b-8192",
messages=prompt,
temperature=0.2,
max_tokens=250
)
output = response.choices[0].message.content.strip()
match = re.search(r'([^\[]+)(\[[^\]]+\])', output)
if match:
return f"{match.group(1).strip()}\n{match.group(2).strip()}", history_log
else:
return f"{output}\n[{english_msg}]", history_log
def ignore_latest():
if history_log:
history_log[-1][4] = "Ignored"
return history_log
def clear_history():
history_log.clear()
detailed_log.clear()
return [], [], "", gr.update(visible=False)
def create_interface():
"""Create the Gradio interface"""
with gr.Blocks(
title="ZeroPhish Gate - Phishing Detection",
theme=gr.themes.Soft(),
css="""
.main-header { text-align: center; margin-bottom: 20px; }
.analysis-output { padding: 15px; border-radius: 10px; }
"""
) as demo:
gr.HTML("""
<div class="main-header">
<h1>π‘οΈ ZeroPhish Gate</h1>
<h3>AI-Powered Phishing & Threat Detection</h3>
<p>Analyze messages, emails, and documents for potential security threats</p>
</div>
""")
# System status
status_msg = "π’ System Ready"
if not (GROQ_SUPPORT and groq_key):
status_msg += " (Advanced AI disabled)"
if not phishing_pipe:
status_msg += " (Using basic detection)"
gr.Markdown(f"**Status:** {status_msg}")
with gr.Row():
with gr.Column(scale=3):
text_input = gr.Textbox(
label="π Message to Analyze",
placeholder="Paste suspicious email, SMS, or message here...",
lines=5
)
file_input = gr.File(
label="π Upload File (Optional)",
file_types=[".txt", ".pdf"] if PDF_SUPPORT else [".txt"]
)
with gr.Column(scale=1):
role = gr.Dropdown(
label="π€ Your Role",
choices=role_choices,
value="Admin"
)
language = gr.Dropdown(
label="π Language",
choices=language_choices,
value="English"
)
analyze_btn = gr.Button(
"π Analyze Message",
variant="primary",
size="lg"
)
new_chat_btn = gr.Button("π New Chat", variant="secondary")
with gr.Row():
output = gr.HTML(
label="π Analysis Results",
elem_classes=["analysis-output"]
)
with gr.Row():
report_btn = gr.Button("π¨ Report to IT", visible=False, variant="stop")
ignore_btn = gr.Button("π Ignore Message", visible=False)
report_msg = gr.Textbox(label="π£ IT Confirmation", visible=False, interactive=False)
with gr.Row():
audio_output = gr.Audio(label="π Voice Output (Click to Play)", interactive=False, autoplay=False)
report_file = gr.File(label="π Download Security Report", interactive=False)
with gr.Accordion("π Risk History & Detailed Reports", open=False):
history_table = gr.Dataframe(
headers=["Time", "Preview", "Threat Score", "Type", "Status"],
label="π Risk History Log",
interactive=False,
wrap=True
)
clear_btn = gr.Button("π§Ή Clear History")
selected_idx = gr.Dropdown(label="π Select Report to View", choices=[], interactive=True)
full_view = gr.Textbox(label="π Detailed Analysis", lines=12, interactive=False)
with gr.Accordion("βΉοΈ Help & Information", open=False):
gr.Markdown("""
### How to Use
1. **Paste or type** the suspicious message in the text box
2. **Upload a file** (PDF or TXT) if needed
3. **Select your role** for personalized advice
4. **Click Analyze** to get results
### Threat Types
- π’ **Safe**: No threats detected
- π‘ **Spam**: Unwanted promotional content
- π **Suspicious**: Potentially harmful content
- π΄ **Phishing**: Attempts to steal information
- π΄ **Malware**: Malicious software threats
### Tips
- Always verify suspicious requests through official channels
- Never click links or download attachments from unknown senders
- When in doubt, contact your IT security team
""")
with gr.Accordion("π Glossary Help", open=False):
gr.Markdown("""
**Hover over underlined blue terms in the analysis to see their definitions:**
- **Phishing**: A type of online scam where attackers trick you into giving away personal information
- **Domain Spoofing**: When a fake website mimics a trusted one by using a similar-looking web address
- **Malware**: Software designed to harm or gain unauthorized access to your device or data
- **Spam**: Unwanted or unsolicited messages, usually advertisements or scams
- **Tone**: The emotional tone in a message, like being urgent or friendly
""")
with gr.Accordion("π€ AI Pipeline Info", open=False):
gr.Markdown("""
**Three-Stage Hybrid Analysis Pipeline:**
1. **Stage 1 - BERT Model:** Technical phishing pattern detection
2. **Stage 1.5 - RAG Reranking:** LLaMA semantic reanalysis for intent understanding
3. **Stage 2 - Final Interpretation:** User-friendly explanation generation
**RAG-Based Reranking Benefits:**
β
**Semantic Understanding:** LLaMA analyzes intent and context, not just patterns
β
**Social Engineering Detection:** Better detection of psychological manipulation
β
**Hybrid Decision Making:** Combines pattern matching with contextual analysis
β
**Reduced False Positives:** More accurate classification of legitimate messages
**How It Works:**
- BERT identifies technical patterns (suspicious links, keywords, etc.)
- LLaMA reanalyzes for social engineering, urgency, and intent
- System combines both analyses for final classification
- Prioritizes safety while reducing false alarms
**Message Classification:**
- Safe: Normal, legitimate messages
- Spam: Unwanted promotional content
- Phishing: Attempts to steal personal information
- Malware: Messages with malicious attachments or links
""")
# Event handlers
analyze_btn.click(
fn=analyze_message_interface,
inputs=[text_input, file_input, language, role],
outputs=[output, report_btn, history_table, selected_idx, full_view, audio_output, report_msg, ignore_btn, report_file]
)
new_chat_btn.click(
fn=new_chat,
inputs=[],
outputs=[text_input, file_input, output, report_btn, ignore_btn, report_msg, audio_output, report_file]
)
selected_idx.change(fn=view_full_report, inputs=[selected_idx], outputs=[full_view])
report_btn.click(fn=report_to_it, inputs=[language], outputs=[report_msg, history_table])
report_btn.click(lambda: gr.update(visible=True), outputs=report_msg)
ignore_btn.click(fn=ignore_latest, outputs=[history_table])
clear_btn.click(fn=clear_history, outputs=[history_table, selected_idx, full_view, report_msg])
return demo
# Create and launch the interface
if __name__ == "__main__":
demo = create_interface()
demo.launch(
server_name="0.0.0.0",
server_port=7860,
show_error=True,
share=False
) |