The field of large language model (LLM) integration is rapidly advancing, with a growing focus on security and reliability. Recent developments have highlighted the importance of addressing vulnerabilities in LLM-based systems, particularly in applications such as robotic systems and tool invocation protocols. Researchers are working to develop unified frameworks that mitigate prompt injection attacks and enforce operational safety, as well as investigating novel attack methods such as parasitic toolchain attacks. Noteworthy papers include: See No Evil, which presents a novel adversarial framework to disrupt the unified referring-matching mechanisms of Referring Multi-Object Tracking models. Enhancing Reliability in LLM-Integrated Robotic Systems proposes a unified framework to mitigate prompt injection attacks and enforce operational safety in LLM-based robotic systems. Exploit Tool Invocation Prompt for Tool Behavior Hijacking in LLM-Based Agentic System reveals TIP-related security risks and proposes defense mechanisms to enhance TIP security. Mind Your Server conducts a systematic study of parasitic toolchain attacks on the MCP ecosystem, revealing a new class of attacks that can hijack entire execution flows.