README.md

---
license: mit
language:
  - en
library_name: pytorch
tags:
  - security
  - cybersecurity
  - network-security
  - c2-detection
  - beacon-detection
  - threat-detection
  - malware-detection
  - logbert
  - transformer
  - safetensors
pipeline_tag: other
---

# C2Sentinel

[![Downloads](https://img.shields.io/badge/dynamic/json?url=https://huggingface.co/api/models/danielostrow/c2sentinel&query=downloads&label=Downloads&color=blue)](https://huggingface.co/danielostrow/c2sentinel)
[![License](https://img.shields.io/badge/License-MIT-green.svg)](https://opensource.org/licenses/MIT)
[![Demo](https://img.shields.io/badge/Demo-Hugging%20Face%20Spaces-yellow)](https://huggingface.co/spaces/danielostrow/c2sentinel)

A machine learning model for detecting Command and Control (C2) beacon communications in network traffic. Built on a fine-tuned [LogBERT](https://arxiv.org/abs/2103.04475) transformer architecture.

**Author:** Daniel Ostrow
**Website:** [neuralintellect.com](https://neuralintellect.com)
**Release Date:** January 18, 2026

---

## Base Model

This model is fine-tuned from the LogBERT architecture for log anomaly detection.

- **Paper:** [LogBERT: Log Anomaly Detection via BERT](https://arxiv.org/abs/2103.04475) (Guo, Yuan, Wu - IJCNN 2021)
- **Original Implementation:** [github.com/HelenGuohx/logbert](https://github.com/HelenGuohx/logbert)

---

## Overview

C2Sentinel analyzes network connection patterns to identify C2 beacon activity. The model uses behavioral analysis rather than port-based filtering, enabling detection of C2 communications on any port. This approach catches C2 activity regardless of whether attackers use expected ports (4444) or attempt to blend in on common ports (443, 80, 53).

### Capabilities

- Detection of 34+ C2 framework behavioral patterns across all ports
- Slow beacon detection (intervals from seconds to hours)
- Legitimate traffic pattern recognition (SSH keepalive, health checks, database connections)
- Optional context enrichment (process information, reputation scores, threat intelligence)
- IP reconnaissance and IOC generation
- Safetensors format for secure model loading

---

## Installation

```bash
pip install torch numpy safetensors huggingface_hub
```

---

## Usage

### Loading from HuggingFace Hub

```python
from c2sentinel import C2Sentinel

sentinel = C2Sentinel.from_pretrained('danielostrow/c2sentinel')
```

### Loading from Local Files

```python
from c2sentinel import C2Sentinel

sentinel = C2Sentinel.load('c2_sentinel')
```

### Analyzing Connections

```python
connections = [
    {
        'timestamp': 1000000,
        'dst_ip': '10.0.0.1',
        'dst_port': 443,
        'bytes_sent': 200,
        'bytes_recv': 500
    },
    {
        'timestamp': 1000060,
        'dst_ip': '10.0.0.1',
        'dst_port': 443,
        'bytes_sent': 200,
        'bytes_recv': 500
    },
]

result = sentinel.analyze(connections)

if result.is_c2:
    print(f"C2 detected: {result.c2_type}")
    print(f"Probability: {result.c2_probability}")
else:
    print("No C2 detected")
```

---

## Connection Record Format

| Field | Type | Required | Description |
|-------|------|----------|-------------|
| `timestamp` | float | Yes | Unix timestamp |
| `dst_ip` | str | Yes | Destination IP address |
| `dst_port` | int | Yes | Destination port |
| `bytes_sent` | int | Yes | Bytes sent |
| `bytes_recv` | int | Yes | Bytes received |
| `src_ip` | str | No | Source IP address |
| `src_port` | int | No | Source port |
| `protocol` | str | No | Protocol (tcp/udp) |
| `duration` | float | No | Connection duration in seconds |

---

## Analysis Options

### Threshold

```python
# Default threshold (0.5)
result = sentinel.analyze(connections)

# Lower threshold for higher sensitivity
result = sentinel.analyze(connections, threshold=0.3)

# Higher threshold for higher precision
result = sentinel.analyze(connections, threshold=0.7)

# Strict mode enforces minimum 0.7 threshold
result = sentinel.analyze(connections, strict_mode=True)
```

### Context

```python
from c2sentinel import ConnectionContext

context = ConnectionContext(
    process_name='sshd',
    known_good=True,
    ip_reputation=0.95,
    dns_queries=['api.example.com']
)

result = sentinel.analyze(connections, context=context)
```

### Whitelist and Blacklist

```python
sentinel.add_whitelist(ips=['8.8.8.8'], domains=['google.com'])
sentinel.add_blacklist(ips=['10.10.10.10'], domains=['malware.example'])
```

---

## Result Object

The `AnalysisResult` object contains:

| Attribute | Type | Description |
|-----------|------|-------------|
| `is_c2` | bool | True if C2 detected |
| `c2_probability` | float | Probability score (0.0-1.0) |
| `c2_type` | str | Detected C2 framework type |
| `confidence` | float | Model confidence |
| `detection_method` | str | Method used (signature/ml/context/whitelist) |
| `immediate_detection` | bool | True if signature-based |
| `risk_factors` | list | Factors supporting C2 classification |
| `mitigating_factors` | list | Factors against C2 classification |
| `matched_legitimate_pattern` | str | Matched legitimate pattern name |
| `service_type` | str | Detected service type |
| `recommendations` | list | Suggested actions |

---

## Batch Analysis

```python
connection_groups = [
    [conn1, conn2, conn3],
    [conn4, conn5, conn6],
]

results = sentinel.analyze_batch(connection_groups)
```

---

## Log File Parsing

```python
with open('conn.log', 'r') as f:
    log_lines = f.readlines()

results = sentinel.analyze_logs(log_lines, group_by_dst=True)
```

Supported formats: JSON, Zeek conn.log, syslog

---

## Reconnaissance

### IP Analysis

```python
info = sentinel.recon.analyze_ip('104.16.132.229')
# Returns: is_valid, is_private, is_cdn, cdn_provider, reverse_dns
```

### Pattern Analysis

```python
patterns = sentinel.recon.analyze_connection_patterns(connections)
# Returns: timing stats, volume stats, behavioral indicators
```

### IOC Generation

```python
if result.is_c2:
    iocs = sentinel.recon.generate_iocs(connections, result.to_dict())
    # Returns: ips, ports, timing_signatures, behavioral_indicators
```

---

## Detection Methodology

### C2 Indicators

- Consistent beacon intervals (low timing variance)
- Consistent packet sizes (low size variance)
- Single persistent destination
- Balanced request/response ratio

### Signature Detection

Immediate detection for high-confidence C2 ports with matching behavioral patterns:
- Port 4444 (Metasploit default)
- Port 5555 (Metasploit alternative)
- Port 31337 (Sliver)
- Port 40056 (Havoc)

### Legitimate Traffic Indicators

- High response size variance
- Asymmetric traffic patterns (small requests, large responses)
- Multiple destinations
- SSH keepalive patterns (small symmetric packets on port 22)
- Health check patterns (regular intervals, variable response sizes)

---

## Model Specifications

| Specification | Value |
|---------------|-------|
| Architecture | LogBERT Transformer |
| Parameters | 4.9 million |
| Feature Dimensions | 40 |
| Encoder Layers | 6 |
| Attention Heads | 8 |
| Hidden Dimension | 256 |
| Format | Safetensors |
| Size | 20 MB |

---

## Repository Contents

| File | Description |
|------|-------------|
| `c2sentinel.py` | Main module with model and analysis code |
| `c2_sentinel.safetensors` | Trained model weights |
| `c2_sentinel.json` | Model configuration |
| `normalization_params.npz` | Feature normalization parameters |
| `API_REFERENCE.md` | Complete API documentation for scripting |
| `LICENSE` | MIT License |

---

## License

MIT License

Copyright (c) 2026 Daniel Ostrow

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.