C2Sentinel

A machine learning model for detecting Command and Control (C2) beacon communications in network traffic. Built on a fine-tuned LogBERT transformer architecture.

Author: Daniel Ostrow Website: 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 (Guo, Yuan, Wu - IJCNN 2021)
• Original Implementation: 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

pip install torch numpy safetensors huggingface_hub

---

Usage

Loading from HuggingFace Hub

from c2sentinel import C2Sentinel

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

Loading from Local Files

from c2sentinel import C2Sentinel

sentinel = C2Sentinel.load('c2_sentinel')

Analyzing Connections

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

# 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

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

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

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

results = sentinel.analyze_batch(connection_groups)

---

Log File Parsing

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

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

Pattern Analysis

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

IOC Generation

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.