The field of AI safety is moving towards a deeper understanding of the vulnerabilities in current safety mechanisms. Recent research has highlighted the limitations of existing defenses against jailbreak attacks, which can be used to exploit large language models and text-to-image systems. The development of new attack methods, such as parallel decoding jailbreaks and persona prompts, has shown that these models can be manipulated to generate harmful content. Furthermore, the use of genetic algorithms and psychological manipulation techniques has demonstrated the potential for self-evolving phishing strategies. The field is also exploring the risks of hateful illusions and the limitations of current content moderation models. Noteworthy papers include: Jailbreaking Large Language Diffusion Models, which presents a novel jailbreak framework for diffusion-based language models. PRISM: Programmatic Reasoning with Image Sequence Manipulation for LVLM Jailbreaking, which proposes a jailbreak framework inspired by Return-Oriented Programming techniques. Anyone Can Jailbreak, which presents a systems-style investigation into prompt-based attacks on LLMs and T2Is. Hate in Plain Sight, which investigates the risks of scalable hateful illusion generation and the potential for bypassing current content moderation models.