Towards trusted AI Week 8 - autonomous vehicles vulnerabilities

Autonomous transportation is one of the most popular research topics, and for good reason

---

EU report warns that AI makes autonomous vehicles highly vulnerable to attack 

Venture Beat, February 22, 2021 For a long time, the issues of the safety of smart machines and trust in them have remained an acute issue in society. The dream of all developers is to achieve such a level of AI, when autonomous vehicles become absolutely reliable and can save lives. However, unfortunately, based on a new study by a new European Union Agency for Cybersecurity (ENISA), modern autonomous vehicles are still very far from this, because, according to researchers, they are susceptible to a wide variety of attacks, including adversarial machine learning attacks, back-end malicious actions, and sensor attacks affecting detection systems. Most frightening is the fact that such attacks can potentially be really dangerous to the health and life of both pedestrians and drivers. It is noted in the report that “the absence of sufficient security knowledge and expertise among developers and system designers on AI cybersecurity is a major barrier that hampers the integration of security in the automotive sector.”

Artificial intelligence (AI) and privacy: 3 key security practices

The Enterprisers Project, February 22, 2021 Before a company implements artificial intelligence technologies, it should familiarize itself with privacy issues and ways to improve it. Here are three basic techniques that can help you a lot in this.

1. Federated learning. It is a machine learning technique aiming to train an algorithm on several decentralized edge devices or servers storing local sample data without sharing it. The technique reduces the risk of a single attack or data breach.
2. Explainable AI (XAI). Here AI is used to enable humans to understand the outcome of a solution. It contrastly differs from the “black box” concept, where even the designers cannot explain why the AI ​​made a particular decision.
3. AIOps/MLOps. The main goal of AIOps and MLOps is to