Ghost trace
Not all cyber attacks use code. Some use people
Created on 8th February 2026
•
Ghost trace
Not all cyber attacks use code. Some use people
The problem Ghost trace solves
Problem GhostTrace Solves:-
People don’t just get hacked through software — they get tricked through messages that manipulate emotions (fear, urgency, authority, trust).
Current tools mostly detect malware or bad links, but they often miss psychological social‑engineering attacks.
GhostTrace fills this gap by analyzing intent + manipulation tactics, not just technical threats.
Challenges we ran into
- Detecting Manipulation Without Over‑Flagging
Problem:
Early versions flagged too many normal messages as “risky” because words like urgent, important, or verify appear in both real and scam messages.
How We Solved It:
Switched from keyword detection → context + intent analysis
Combined signals like:
Emotional pressure
Authority impersonation
Urgency + action demand together
Added confidence scoring instead of simple yes/no flag
2. Speed vs Accuracy Tradeoff
Problem:
Deep analysis improved accuracy but made response time slower — bad for real‑time chat scanning.
How We Solved It:
Built a 2‑stage pipeline:
-
Fast lightweight risk pre‑check
-
Deep analysis only if needed
This kept responses fast while staying accurate -
Handling Different Message Formats
Problem:
Messages come in many forms:
Emails
SMS
Social media chats
Mixed text + links
Each has different structure.
How We Solved It:
Built a message normalization layer
Converts everything into a standard format before analysis
- Separating “Suspicious Tone” From Normal Human Emotion
Problem:
People naturally use emotion in normal messages (e.g., “Please do this ASAP”).
We needed to detect manipulation, not just emotion.
How We Solved It:
Trained models to look for:
Forced action + consequence threat
Fake authority signals
Pressure + secrecy combos
Not just emotional words alone.
Tracks Applied (1)
Google Gemini
Major League Hacking
Technologies used