The field of GUI agents and computer-use security is rapidly advancing, with a focus on developing more robust, efficient, and secure systems. Researchers are exploring new methods for training and evaluating GUI agents, including the use of multimodal large language models and structured exploration of web environments. Additionally, there is a growing concern about the security risks associated with computer-use agents, particularly in regards to visual prompt injection attacks and deception attacks. To address these risks, researchers are proposing new defense mechanisms and evaluation benchmarks. Notable papers in this area include VPI-Bench, which introduces a benchmark for evaluating agent robustness under visual prompt injection threats, and GUI-Actor, which proposes a coordinate-free visual grounding method for GUI agents. Other noteworthy papers include Surfer-H, which presents a cost-efficient web agent powered by open weights, and macOSWorld, which introduces a comprehensive benchmark for evaluating GUI agents on macOS.