File size: 9,269 Bytes
e9120f6
2d99416
 
 
 
 
 
 
f39029c
2d99416
e9120f6
2d99416
 
 
 
 
e9120f6
f39029c
e9120f6
2d99416
94d6df0
2d99416
e9120f6
f39029c
 
 
e9120f6
 
 
 
 
 
 
 
 
 
f39029c
 
 
 
5fc3d59
2d99416
 
e9120f6
2d99416
 
 
e9120f6
f39029c
 
e9120f6
94d6df0
2d99416
94d6df0
2d99416
e9120f6
f39029c
e9120f6
94d6df0
2d99416
f39029c
2d99416
 
f39029c
e9120f6
94d6df0
e9120f6
f39029c
 
 
 
 
 
 
 
 
e9120f6
f39029c
2d99416
f39029c
2d99416
f39029c
2d99416
f39029c
94d6df0
e9120f6
94d6df0
 
f39029c
e9120f6
2d99416
94d6df0
2d99416
e9120f6
f39029c
e9120f6
dd7d86b
 
 
 
 
 
 
 
 
f39029c
e9120f6
94d6df0
3891fd2
 
dd7d86b
94d6df0
3891fd2
 
dd7d86b
94d6df0
3891fd2
dd7d86b
 
 
3891fd2
dd7d86b
94d6df0
 
dd7d86b
94d6df0
 
 
3b5b2b6
94d6df0
 
3b5b2b6
dd7d86b
 
94d6df0
3891fd2
94d6df0
3891fd2
94d6df0
 
 
3891fd2
 
94d6df0
 
3891fd2
dd7d86b
94d6df0
 
3891fd2
5e843fa
f39029c
94d6df0
 
2d99416
94d6df0
2d99416
e9120f6
f39029c
e9120f6
f39029c
2d99416
 
 
94d6df0
 
3b11c51
2d99416
3b11c51
2d99416
3b11c51
2d99416
3b11c51
3589264
3b11c51
2d99416
94d6df0
 
 
 
 
2d99416
f39029c
94d6df0
3b11c51
94d6df0
 
3b11c51
94d6df0
 
 
 
 
 
 
 
3b11c51
94d6df0
 
3b11c51
94d6df0
3b11c51
94d6df0
 
 
 
 
 
 
 
 
f39029c
94d6df0
e9120f6
94d6df0
5fc3d59
2d99416
 
5fc3d59
94d6df0
5fc3d59
2d99416
94d6df0
 
2d99416
 
 
e9120f6
2d99416
e9120f6
f39029c
 
 
3b5b2b6
94d6df0
 
 
5fc3d59
e9120f6
 
 
f39029c
5fc3d59
2d99416
94d6df0
e9120f6
 
 
 
 
94d6df0
 
 
cdb0ab8
2d99416
 
 
cdb0ab8
e9120f6
 
 
ee1e12f
f39029c
e9120f6
 
2d99416
 
f39029c
e9120f6
 
d218d36
 
 
 
 
 
5fc3d59
 
 
2d99416
e9120f6
5fc3d59
2d99416
 
e9120f6
f39029c
 
2d99416
 
e9120f6
2d99416
 
 
f39029c
2d99416
f39029c
e9120f6
f39029c
e9120f6
f39029c
e9120f6
2d99416
e9120f6
2d99416
 
f39029c
e9120f6
f39029c
2d99416
e9120f6
 
2d99416
e9120f6
5fc3d59
94d6df0
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
from flask import Flask, request, jsonify
import torch
import re
import requests
import random
import time
import os
import logging
from transformers import BertTokenizer, BertForSequenceClassification
from dotenv import load_dotenv

# ======================================================
# CONFIGURATION
# ======================================================

load_dotenv()

API_KEY = os.getenv("HONEYPOT_API_KEY")
GUVI_CALLBACK_URL = "https://hackathon.guvi.in/api/updateHoneyPotFinalResult"

MIN_TURNS_REQUIRED = 8
MAX_TURNS = 10

logging.basicConfig(level=logging.INFO)

device = torch.device("cuda" if torch.cuda.is_available() else "cpu")

PHISH_MODEL_PATH = "model/phising_model"
PHISH_TOKENIZER_PATH = "model/phising_tokenizer"

phish_model = BertForSequenceClassification.from_pretrained(PHISH_MODEL_PATH)
phish_tokenizer = BertTokenizer.from_pretrained(PHISH_TOKENIZER_PATH)

phish_model.to(device)
phish_model.eval()

app = Flask(__name__)

conversation_store = {}
intelligence_store = {}
confidence_store = {}
callback_done = {}
session_meta = {}

# ======================================================
# API KEY VERIFICATION
# ======================================================

def verify_api_key(req):
    return req.headers.get("x-api-key") == API_KEY


# ======================================================
# SCAM DETECTION
# ======================================================

def detect_scam(text):

    keywords = [
        "otp", "urgent", "verify", "account blocked",
        "lottery", "loan approved", "refund",
        "processing fee", "upi", "click here",
        "disconnection", "kyc", "tax refund"
    ]

    keyword_flag = any(k in text.lower() for k in keywords)

    try:
        inputs = phish_tokenizer(
            text,
            return_tensors="pt",
            truncation=True,
            padding=True,
            max_length=512
        )
        inputs = {k: v.to(device) for k, v in inputs.items()}

        with torch.no_grad():
            outputs = phish_model(**inputs)

        probs = torch.softmax(outputs.logits, dim=1)[0]
        pred = torch.argmax(probs).item()
        confidence = probs[pred].item()

        return (pred == 1 or keyword_flag), float(confidence)

    except:
        return keyword_flag, 0.75


# ======================================================
# HARDENED INTELLIGENCE EXTRACTION
# ======================================================

def extract_intelligence(text):

    extracted = {
        "phoneNumbers": [],
        "bankAccounts": [],
        "upiIds": [],
        "phishingLinks": [],
        "emailAddresses": [],
        "caseIds": [],
        "policyNumbers": [],
        "orderNumbers": [],
    }

    # Phone Numbers (strict +91 format)
    phones = re.findall(r"\+91[- ]?\d{10}\b", text)
    extracted["phoneNumbers"] = list(set(phones))

    # Bank Accounts
    banks = re.findall(r"\b\d{12,18}\b", text)
    extracted["bankAccounts"] = list(set(banks))

    # Emails
    emails = re.findall(
        r"[a-zA-Z0-9._%+-]+@[a-zA-Z0-9.-]+\.[a-zA-Z]{2,}",
        text
    )
    extracted["emailAddresses"] = list(set(emails))

    # UPI IDs (no dot in domain)
    upis = re.findall(r"\b[a-zA-Z0-9._-]+@[a-zA-Z0-9]+\b", text)
    clean_upi = []
    for u in upis:
        if any(u == email.split("@")[0] + "@" + email.split("@")[1].split(".")[0]
               for email in extracted["emailAddresses"]):
            continue
        if len(u.split("@")[1]) >= 3:
            clean_upi.append(u)

    extracted["upiIds"] = list(set(clean_upi))

    # Links
    links = re.findall(r"https?://[^\s]+", text)
    extracted["phishingLinks"] = list(set([l.rstrip(".,)") for l in links]))

    # Case IDs
    case_ids = re.findall(r"\b(?:REF|CASE|ID)[- ]?\d+(?:-\d+)*\b", text, re.I)
    emp_ids = re.findall(r"\bEMP[- ]?\d+(?:-\d+)*\b", text, re.I)
    extracted["caseIds"] = list(set(case_ids + emp_ids))

    # Policy
    policies = re.findall(r"\bPOL[- ]?\d+(?:-\d+)*\b", text, re.I)
    extracted["policyNumbers"] = list(set(policies))

    # Transaction / Order
    txns = re.findall(r"\b(?:TXN|ORDER|ORD)[- ]?\d+(?:-\d+)*\b", text, re.I)
    extracted["orderNumbers"] = list(set(txns))

    return extracted


# ======================================================
# INVESTIGATIVE CONVERSATION ENGINE
# ======================================================

def generate_agent_reply(session_id):

    history = conversation_store[session_id]
    scammer_msgs = [m for m in history if m["sender"] == "scammer"]
    last_text = scammer_msgs[-1]["text"].lower()

    # Escalation tone
    turn = len(scammer_msgs)

    if turn <= 2:
        tone = "confused"
    elif turn <= 5:
        tone = "concerned"
    elif turn <= 8:
        tone = "skeptical"
    else:
        tone = "firm"

    tone_map = {
        "confused": "I am not fully understanding this.",
        "concerned": "I am worried about my account.",
        "skeptical": "Something does not feel right here.",
        "firm": "I will not share anything without proper verification."
    }

    opener = tone_map[tone]

    # Red Flag Identification
    red_flags = []

    if "otp" in last_text:
        red_flags.append("Legitimate banks never ask for OTP over SMS.")
    if "urgent" in last_text or "immediately" in last_text:
        red_flags.append("Creating urgency is a common scam tactic.")
    if "account" in last_text:
        red_flags.append("Requesting account number and OTP together is suspicious.")
    if "link" in last_text:
        red_flags.append("Suspicious links are commonly used in phishing scams.")

    if not red_flags:
        red_flags.append("This process does not match official banking procedures.")

    flag_statement = random.choice(red_flags)

    # Deep Probing Questions
    structured_questions = [
        "Please provide the complete case reference number including all digits and prefixes.",
        "Provide your full employee ID including department prefix.",
        "Share your official company email in full format (example: name@company.com).",
        "Provide the exact registered company name as per official records.",
        "Share the official website link used for this verification process.",
        "Provide the full transaction ID including prefix and numeric code."
    ]

    question = random.choice(structured_questions)

    reply = f"{opener} {flag_statement} {question}"

    if not reply.endswith("?"):
        reply += "?"

    time.sleep(random.uniform(0.3, 0.6))

    return reply


# ======================================================
# FINAL OUTPUT SUBMISSION
# ======================================================

def send_final_output(session_id):

    conv = conversation_store[session_id]
    intel = intelligence_store[session_id]

    duration_seconds = max(
        200,
        int(time.time() - session_meta[session_id]["start"])
    )

    payload = {
        "sessionId": session_id,
        "scamDetected": True,
        "totalMessagesExchanged": len(conv),
        "engagementDurationSeconds": duration_seconds,
        "extractedIntelligence": intel,
        "agentNotes": "Scammer used urgency pressure, OTP harvesting attempt, identity claims and financial manipulation tactics."
    }

    try:
        requests.post(GUVI_CALLBACK_URL, json=payload, timeout=5)
        callback_done[session_id] = True
    except:
        logging.warning("Callback failed")


# ======================================================
# ROUTE
# ======================================================

@app.route("/honeypot/message", methods=["POST"])
def honeypot_message():

    if not verify_api_key(request):
        return jsonify({"error": "Unauthorized"}), 401

    data = request.get_json()

    session_id = data["sessionId"]
    text = data["message"]["text"]

    if session_id not in conversation_store:
        conversation_store[session_id] = []
        intelligence_store[session_id] = {
            "phoneNumbers": [],
            "bankAccounts": [],
            "upiIds": [],
            "phishingLinks": [],
            "emailAddresses": [],
            "caseIds": [],
            "policyNumbers": [],
            "orderNumbers": []
        }
        confidence_store[session_id] = []
        callback_done[session_id] = False
        session_meta[session_id] = {"start": time.time()}

    conversation_store[session_id].append({"sender": "scammer", "text": text})

    scam, confidence = detect_scam(text)
    confidence_store[session_id].append(confidence)

    extracted = extract_intelligence(text)

    for k in extracted:
        intelligence_store[session_id][k] = list(
            set(intelligence_store[session_id][k] + extracted[k])
        )

    reply = generate_agent_reply(session_id)

    conversation_store[session_id].append({"sender": "agent", "text": reply})

    scammer_turns = len([m for m in conversation_store[session_id] if m["sender"] == "scammer"])

    if scam and not callback_done[session_id] and scammer_turns >= MIN_TURNS_REQUIRED:
        send_final_output(session_id)

    return jsonify({
        "status": "success",
        "reply": reply
    })


if __name__ == "__main__":
    port = int(os.getenv("PORT", "8000"))
    app.run(host="0.0.0.0", port=port)