🛡️ Autobook - ML Security Incident Response Runbooks Automation

Executable Jupyter notebook runbooks for detecting, investigating, and responding to security incidents in ML/AI infrastructure.

📋 Overview

This repository contains Security incident response runbooks designed for security operations teams protecting ML systems. Each runbook is an interactive Jupyter notebook that guides security engineers through the complete incident response lifecycle.

Key Features

• Executable Investigation - Run analysis code directly in notebooks
• GCP Integration - Built-in support for Cloud Logging, Vertex AI, BigQuery, KMS
• MITRE Mapping - Aligned to ATT&CK and ATLAS frameworks
• Evidence Collection - SHA-256 hashed evidence packages for forensics
• Approval Gates - Containment actions require explicit approval

📚 Runbooks

#	Runbook	Severity	MITRE Reference
01	ML Model Exfiltration from GCS	🔴 HIGH	T1530, T1567.002
02	ML Training Job Manipulation	🔴 CRITICAL	T1565.001
03	ML Inference API Abuse	🔴 HIGH	AML.T0024
04	K8s ML Workload Compromise	🔴 CRITICAL	T1611
05	ML Pipeline Code Injection	🔴 HIGH	T1195.002
06	AR Device Firmware Tampering	🔴 CRITICAL	T1195.002, T1542
07	ML Training Data Poisoning	🔴 HIGH	AML.T0020
08	ML Model Registry Tampering	🔴 CRITICAL	T1195.002, T1565.001
09	Adversarial ML Input Detection	🟡 MEDIUM	AML.T0015, AML.T0016
10	ML Secrets Exposure	🔴 HIGH	T1552.001, T1552.004

🏗️ Runbook Structure

Each runbook follows a standardized 7-section incident response workflow:

1️⃣ Initial Triage & Alert Validation
   └── Configure alert parameters, validate detection

2️⃣ Scope Assessment
   └── Identify all affected resources and blast radius

3️⃣ Detailed Analysis
   └── Deep investigation with audit log queries

4️⃣ Impact Assessment
   └── Determine business impact and risk level

5️⃣ Containment Actions ⚠️
   └── Approval-gated response actions

6️⃣ Evidence Collection
   └── Forensic evidence with integrity hashing

7️⃣ Remediation Checklist
   └── Recovery procedures and hardening steps

🚀 Quick Start

Prerequisites

# Required Python packages
pip install google-cloud-storage google-cloud-logging google-cloud-aiplatform \
            google-cloud-bigquery google-cloud-kms pandas numpy

GCP Authentication

# Authenticate with GCP
gcloud auth application-default login

# Set project
gcloud config set project YOUR_PROJECT_ID

Using a Runbook

1. Open the relevant runbook in Jupyter/VS Code
2. Update the Configuration cell with alert details:

   ALERT_TIMESTAMP = "2024-01-15T10:30:00Z"
   PROJECT_ID = "your-gcp-project-id"
   # ... other alert-specific parameters

3. Execute cells sequentially through investigation phases
4. Set CONTAINMENT_APPROVED = True only after manager approval
5. Export evidence package to secure storage

🔧 Configuration

Required GCP APIs

Enable these APIs in your project:

• Cloud Logging API
• Cloud Storage API
• Vertex AI API
• BigQuery API
• Cloud KMS API
• IAM API

Required IAM Roles

The service account running these runbooks needs:

• roles/logging.viewer - Read audit logs
• roles/storage.objectViewer - Access GCS buckets
• roles/aiplatform.viewer - View Vertex AI resources
• roles/bigquery.dataViewer - Query BigQuery
• roles/iam.securityReviewer - Review IAM policies

Evidence Storage

Create an evidence bucket with restricted access:

gsutil mb -l us-central1 gs://security-incident-evidence
gsutil iam ch serviceAccount:YOUR_SA@PROJECT.iam.gserviceaccount.com:objectCreator \
    gs://security-incident-evidence

📊 MITRE Framework Mapping

ATT&CK Techniques Covered

Technique ID	Name	Runbooks
T1530	Data from Cloud Storage	01
T1565.001	Stored Data Manipulation	02, 08
T1611	Escape to Host	04
T1195.002	Compromise Software Supply Chain	05, 06, 08
T1542	Pre-OS Boot	06
T1552.001	Credentials In Files	10
T1552.004	Private Keys	10
T1567.002	Exfiltration to Cloud Storage	01

ATLAS Techniques Covered

Technique ID	Name	Runbooks
AML.T0015	Evade ML Model	09
AML.T0016	Obtain Capabilities	09
AML.T0020	Poison Training Data	07
AML.T0024	Exfiltration via ML Inference API	03

🔗 References

• MITRE ATT&CK
• MITRE ATLAS
• GCP Security Best Practices
• NIST Incident Response Guide

---

⚠️ Disclaimer: These runbooks are templates. Customize containment actions and thresholds for your environment before production use.