OTArmor: Securing Automotive Over-the-Air Updates Against Malware Using Generative Modeling
Bibliographic record
Abstract
The rapid rise of connected autonomous vehicles (CAVs) demands a new level of cybersecurity vigilance, particularly when it comes to over-the-air (OTA) updates. While OTA updates are essential for keeping vehicles secure and up-to-date, they also expand the attack surface, creating new opportunities for malicious actors to exploit. Infotainment systems, often overlooked in favor of safety-critical electronic control units (ECUs), represent a particularly vulnerable entry point. Traditional signature-based malware detection methods fall short in this domain, especially against zero-day threats, underscoring the urgent need for more adaptive and resilient defenses. To close this gap, we introduce OTArmor, a generative AI (GenAI)–driven defense mechanism for in-vehicle infotainment (IVI) systems. At its core is a variational autoencoder (VAE) that learns the distribution of benign OTA updates and flags deviations as potential malware, providing a decisive advantage over conventional supervised models that depend on labeled attack data and struggle to adapt to novel or evolving threats. Unlike existing solutions focused on ECUs and trained on broad, generalized datasets, OTArmor is specifically tailored to the IVI environment and trained exclusively on real OTA update data. Before installation, OTArmor screens update artifacts by analyzing file size, entropy, byte occurrence, and n-gram features. We benchmarked its performance against Principal Component Analysis (PCA) and Isolation Forest, testing across contamination rates of 10%, 20%, 50%, and 100%. The results demonstrate that our OTArmor approach consistently identifies malicious files with ROC–AUC values of nearly 0.992 and F1-score of up to 0.985, substantially outperforming PCA and Isolation Forest. To our knowledge, this is the first study to leverage real OTA update data to replicate a production-grade IVI environment for malware detection. By combining generative modeling with domain-specific training, OTArmor provides a practical and forward-looking defense against emerging threats in IVI systems.
Fetched live from OpenAlex and de-inverted. Abstracts are not stored in this database: the inverted indexes are 8.6 GB of the frame’s 9.3 GB of text, and the host has 13 GB free.
How this classification was reachedexpand
Full frame distilled prediction
Teacher imitationNot calibrated prevalence, not ground truth. Human validation pending. Learned from the 10,348 direct Codex labels and 10,348 direct Gemma labels. Candidate is the union of thresholded teacher heads; consensus is their intersection. These outputs are machine_predicted_unvalidated and are not human labels or direct frontier model labels.
Codex and Gemma teacher scores by category
| Category | Codex | Gemma |
|---|---|---|
| Metaresearch | 0.001 | 0.000 |
| Meta-epidemiology (narrow) | 0.001 | 0.001 |
| Meta-epidemiology (broad) | 0.001 | 0.000 |
| Bibliometrics | 0.001 | 0.003 |
| Science and technology studies | 0.002 | 0.000 |
| Scholarly communication | 0.001 | 0.002 |
| Open science | 0.002 | 0.002 |
| Research integrity | 0.000 | 0.001 |
| Insufficient payload (model declined to judge) | 0.000 | 0.000 |
Machine scores (provisional)
The two teacher heads of the student model, read on this work. A score orders the frame for review; it never asserts a category, and the validation status ships verbatim with every row.
Baseline scores from an immature model (maturity gate not passed, 7 training rounds). Scores rank; they never assert a category.
score_only:v0-immature-baseline · verbatim from the scoring run: score_only means the number may rank works, and no category label ships from itClassification
machine, unvalidatedMachine predicted; a candidate call from one teacher head, not a consensus.
How this classification was reached, model by model and score by score, is at the end of the page under "How this classification was reached".