Automotive AI Security Audit
A vehicle's AI makes safety-critical decisions in real time, with no retry. An audit tests perception, sensor fusion, software updates and in-car assistants against deliberate attackers, not only against bad weather.
See the workflow →Four Areas to Test
A sector audit covers each of these areas.
- Perception & sensors
- Fusion & planning
- Updates & connectivity
- Cabin assistants
Core Concepts
The foundations of this sector's AI risk.
Why cars are a special case
A wrong output is physical. The pipeline runs from sensors to actuation, and an attacker only needs to fool one stage.
- Camera: adversarial patches and stickers on signs or road markings
- LIDAR/RADAR: spoofing and injection of phantom objects
- Whole fleet: one compromised OTA update reaches every vehicle
Regulatory frame
Type-approval and safety standards already apply; AI-specific requirements are being added on top.
- UN R155 (cybersecurity management) and UN R156 (software updates), with ISO/SAE 21434
- ISO 26262 (functional safety), ISO 21448 (SOTIF) and ISO/PAS 8800 (AI safety in road vehicles)
- EU AI Act: vehicles fall under type-approval legislation, so AI requirements arrive through that route
The Audit Workflow
Each phase is methodical and repeatable.
Map the pipeline
List sensors, ECUs, CAN/Ethernet buses, embedded models and every external interface (OTA, V2X, Bluetooth, apps).
Test perception and fusion
Run adversarial patches, noise, lighting and occlusion tests in simulation; then inject conflicting LIDAR, RADAR and camera data and check which sensor wins.
Attack updates and interfaces
Check update signing, rollback protection and model integrity; fuzz diagnostic and telematics interfaces; test the voice assistant for injected commands.
Report against safety cases
Rate findings by safety impact, map them to R155 and ISO 21434 risk assessments, and agree remediation with the OEM or supplier.
Related Pages
Deep dives into complementary topics.
Multimodal & Vision Security
Physical adversarial attacks.
AI Supply Chain Security
Datasets, models, dependencies.
AI Agent Security Audit
Tool use, privileges, orchestration.
AI Governance & EU AI Act
Obligations, controls, audit evidence.