The field of autonomous driving is rapidly evolving, with a growing focus on ensuring the safety and security of these systems. Recent research has highlighted the vulnerabilities of autonomous driving models to adversarial attacks, which can compromise their ability to make accurate decisions. In response, researchers are exploring new methods to improve the robustness and reliability of these models, including the use of large language models and multimodal safety alignment frameworks. Notably, some studies have demonstrated the effectiveness of physical adversarial attacks against stereo-based binocular depth estimation and traffic sign recognition systems. To address these challenges, researchers are developing novel defense techniques, such as multi-layered response filtering and rule-governed policy optimization. Overall, the field is moving towards a more comprehensive understanding of the safety and security risks associated with autonomous driving, and the development of innovative solutions to mitigate these risks. Noteworthy papers include: Are LLMs The Way Forward, which investigates the use of large language models for autonomous highway driving and highlights their limitations in safety-critical control tasks. T2I-Based Physical-World Appearance Attack against Traffic Sign Recognition Systems, which presents a novel framework for generating physically robust and effective appearance attacks against traffic sign recognition systems. SafeGRPO, which proposes a self-rewarded multimodal safety alignment framework to improve the safety awareness of multimodal large language models. Robustness of LLM-enabled vehicle trajectory prediction, which analyzes the vulnerability of LLM-based prediction models to adversarial manipulation and highlights the need for robustness-oriented design. Jailbreaking Large Vision Language Models, which systematically analyzes the vulnerabilities of large vision language models to jailbreaking attacks and proposes a multi-layered response filtering defense technique. Cheating Stereo Matching, which proposes a texture-enabled physical adversarial attack against stereo matching models and demonstrates its effectiveness in fooling these models into producing erroneous depth information. Attacking Autonomous Driving Agents, which evaluates the risk of adversarial examples to autonomous driving agents and highlights the importance of considering the entire driving system, rather than just individual machine learning models.