VehicleSec 26 - WIP: On the Real-World Applicability of Automotive CAN Intrusion Detection Systems @UsenixOrg
VehicleSec 26 - WIP: On the Real-World Applicability of Automotive CAN Intrusion Detection Systems  @UsenixOrg
Uploaded September 2026 | Updated November 2026, 5 days ago
WIP: On the Real-World Applicability of Automotive CAN Intrusion Detection Systems

Kristijan Lazeski, Darmstadt University of Applied Sciences and European University of Technology; Michael Zohner, University of Applied Sciences Fulda; Jannis Hamborg, Darmstadt University of Applied Sciences and European University of Technology; Christoph Krauß and Alexander Wiesmaier, Darmstadt University of Applied Sciences, European University of Technology, and National Research Center for Applied Cybersecurity ATHENE

Machine Learning (ML)-based Controller Area Network (CAN) Intrusion Detection Systems (IDS) are widely studied and used to detect and mitigate attacks on in-vehicle communication. Despite frequent reports on high detection performance for well-established datasets, the practical applicability and robustness against real-world attackers of the trained solutions is rarely discussed. This work presents a preliminary analysis of the practical robustness of ML-based CAN IDS solutions, with a focus on per-frame, binary intrusion detection approaches. Our approach consists of three steps: (1) We analyze common datasets used for training and evaluating ML-based CAN IDS, (2) train human-interpretable decision trees on the datasets and investigate their learned classification rules1, and (3) critically evaluate the models against a practical attacker model. We observe that many commonly used CAN IDS datasets only include simple and easily distinguishable attacks. Consequently, ML-based models trained on these datasets tend to learn dataset-specific statistical artifacts or shortcuts rather than robust or semantically meaningful detection logic, a finding we demonstrate specifically within the scope of per-frame, binary detection. While our results are preliminary and limited to this detection paradigm, they show that high detection performance is not inherently indicative of practical robustness. We therefore argue that new detection strategies founded upon vehicle specifications and domain knowledge, in combination with ML tools, may offer more transparent and reliable protection than purely ML-driven approaches, and we call for broader evaluation across more diverse detection settings in future work.

View the full VehicleSec '26 program at usenix.org/conference/vehiclesec26/program
VehicleSec 26 - WIP: On the Real-World Applicability of Automotive CAN Intrusion Detection SystemsWOOT 26 - Security Analysis of LTE Connectivity in Connected Cars: A Case Study of TeslaWOOT 26 - SoK: 20 Years of Power, Privilege, and Peril in x86 System Management ModeSREcon26 Europe/Middle East/Africa - How Uber Validates Failover Resilience in ProductionUSENIX Security 26 - Exploring Privacy Negotiation Strategies for Camera-Equipped Smart Home Devic…USENIX Security 26 - PICS: Private Intersection over Committed (and reusable) SetsSREcon26 Europe/Middle East/Africa - Safe, Empowered, Tired: The Human Side of 24x7 Banking SRE...WOOT 26 - Swarm in EM Hay: Particle Swarm-guided Probe Placement for EM SCAWOOT 26 - SoK: The Constant Time ModelUSENIX Security 26 - End-to-End Encrypted Collaborative DocumentsUSENIX Security 26 - Perils of Parallelism: Transaction Fee Mechanisms under Execution UncertaintyNSDI 26 - Net-P4ct: Enhanced WAN Bandwidth Fair Sharing Using P4 Programmable Switches
USENIX |

VehicleSec '26 - WIP: On the Real-World Applicability of Automotive CAN Intrusion Detection Systems

SHARE TO X SHARE TO REDDIT SHARE TO FACEBOOK WALLPAPER