Educational and defensive security project.
This repository contains a completely local synthetic security incident simulation.
It does not connect to, scan, probe, or attack Binance or any external infrastructure.
- Synthetic security event generation
- Suspicious authentication detection
- Privilege-context anomaly detection
- Simulated container-access alerts
- Simulated transaction anomaly detection
- Automated Markdown incident reports
python -m venv .venv source .venv/bin/activate ./run.sh
All events and incidents in this project are synthetic and local.
Standalone, stdlib-only simulator of the reported Oct-2026 Binance chain: compromised DevOps laptop -> session-token reuse -> service-container login -> signer abuse (burst withdrawals + EIP-7702-style batch authorizations).
Timeline: ~36h sparse preparation, ~20min dense execution.
ChainDetector fires on the correlated sequence, not isolated events.
Measured on the default seed (7):
| Alert | Latency from token reuse | Funds in flight |
|---|---|---|
| SIGNER_ABUSE_CHAIN #1 | t+5.1 min | $33M |
| EIP7702_BATCH | t+7.5 min | $46M |
| SIGNER_ABUSE_CHAIN #2 | t+9.6 min | $60M |
| PREP_THEN_EXECUTE | t+0.0 min (early warning) | $0 |
Run:
python realistic_hijack.py --json events_realistic.json
python detector/analyze_realistic.py events_realistic.json
Alert count is stable across seeds 1, 7, 42, 99, 1234 (4 alerts each), so the
detector is not tuned to a single seed. The sweep in sweep.py confirms no
false positives on the 400-event baseline across window_min 10–60 and
batch_n 10–40.