Attacking edge AI
Edge AI refers to the deployment of artificial intelligence algorithms on local hardware devices close to the data source rather than relying on a centralized AI service such as the one we used in ImReCs. This approach allows for faster real-time data processing, reduced latency, and lower bandwidth usage. Examples of this include mobile apps, autonomous cars, drones, and IoT devices, such as security cameras or other smart home devices.
However, edge AI presents several security challenges, especially in the context of model integrity attacks. The distributed nature of edge AI makes these kinds of systems more vulnerable to physical and cyberattacks, and ensuring data privacy and integrity becomes more complex, primarily when the devices are operating in unsecured environments.
Important note
The research discussed in the neural payload injection-based paper (referenced in the previous section) covered 116 apps on Google Play, using a model on a device....
