Automating Vehicle SOA Threat Analysis Using a Model-Based Methodology

Yuri Gil Dantas, Simon Barner, Pei Ke, Vivek Nigam, Ulrich SchöppICISSP 2023 · DOI

TL;DR

This paper introduces a novel model-based methodology and an automated toolchain (LAUFEN) to systematically identify potential cyber-attacks against autonomous vehicles that could compromise safety. By integrating safety analysis (like STPA) with a formal intruder model and system architecture, they can automatically discover attack paths, including previously unreported ones, to enhance the security of Service-Oriented Architectures (SOA) in vehicles.

What Problem Does This Paper Address?

The increasing complexity of autonomous vehicles, driven by the adoption of Service-Oriented Architectures (SOA), introduces significant cybersecurity challenges. While SOA facilitates advanced features like autonomous driving and over-the-air updates, it also expands the attack surface, potentially leading to safety-critical incidents. Current industry standards (e.g., ISO 21434) mandate thorough threat analysis, but manual methods are often time-consuming, error-prone, and lack rigorous traceability between safety and security concerns, especially when systems are updated. This gap leaves a critical need for automated, systematic, and traceable methods to identify and mitigate threats that could impact road-user safety.

Methods

The core of this work is a model-based engineering methodology that bridges safety and security analysis. It starts with a formal model of the vehicle's SOA architecture, including its logical and physical components. This model is then used in conjunction with safety analysis techniques like Hazard Analysis and Risk Assessment (HARA) and System Theoretic Process Analysis (STPA) to identify hazards and loss scenarios.

Key to their approach is the "Safety to Security" principle, where safety artifacts (like loss scenarios) are used as input for security analysis. They formalize an "Intruder Model" that captures the capabilities of both external and internal attackers within an SOA context. Finally, they developed a tool called LAUFEN (vehicle threAt analysis aUtomation For sErvice-oriented architectures), which leverages logic programming to automate the enumeration of attack paths based on the system model, intruder model, and identified safety-related damage/threat scenarios. This automation aims to provide precise guarantees and traceability between safety goals and security mitigations.

Datasets

Not discussed in the paper. The paper focuses on modeling an existing system (Apollo) rather than training on a dataset.

Key Insights and Ideas

  • Bridging Safety and Security is Crucial: The paper strongly advocates for integrating safety and security analysis from the outset, using safety artifacts as a foundation for security assessments. This ensures that security efforts are directly aligned with preventing safety-critical failures.
  • Automation is Key for Complexity: The sheer complexity of modern automotive SOA necessitates automated tools. Manual threat analysis is insufficient to cover all potential attack vectors, especially in dynamic environments with frequent software updates.
  • Formal Intruder Models Enable Precise Analysis: By formalizing an intruder model, the authors can systematically explore how attackers might exploit SOA vulnerabilities, leading to more comprehensive attack path discovery.
  • Traceability is Paramount: The model-based approach provides a clear link between system design, safety hazards, and identified security threats, which is essential for assurance cases and regulatory compliance.
  • Unexpected Attacks Can Be Found: The automated enumeration of attack paths, even on well-studied systems like Apollo, can uncover vulnerabilities that might have been overlooked by manual analysis.

Key Results

The authors successfully applied their methodology and the LAUFEN tool to a faithful model of the Apollo autonomous driving stack. They identified a significant number of attack paths:

  • Outsider Intruder: 152 attack paths identified.
  • Insider Intruder: 94 attack paths identified.

The execution time for LAUFEN to compute these paths was remarkably fast, taking only 1.11 seconds for the outsider intruder and 0.06 seconds for the insider intruder. This demonstrates the scalability and efficiency of their automated approach. Notably, their analysis uncovered attack paths that had not been previously reported in the literature, highlighting the value of their systematic methodology. They also provided examples of specific attacks, such as V2X traffic light overwrite and route/mapping injection, and discussed potential countermeasures like digital signatures and firewalls.

Limitations and Future Work

The authors acknowledge that while their approach provides a comprehensive set of attack paths, the behavior specification of the system could lead to even more accurate information regarding assets and potential attacks. They suggest that future work could involve incorporating behavior specifications into their Apollo model and using advanced reasoning tools to verify security properties of SOA protocols. Additionally, they propose combining their attack path identification methods with formal behavior-based reasoning to achieve both the precision of formal methods and the scalability of their current approach. They also note that while their method identifies potential attacks, the actual implementation and effectiveness of countermeasures require further investigation.

Summary updated

BibTeX
@inproceedings{dantas2023automating,
  author = {Yuri Gil Dantas and Simon Barner and Pei Ke and Vivek Nigam and Ulrich Schöpp},
  title = {{Automating Vehicle SOA Threat Analysis Using a Model-Based Methodology}},
  booktitle = {ICISSP},
  year = {2023},
  doi = {10.5220/0011786400003405},
}
Figure