Researchers introduce PlanFlip, a framework to attack multi-agent LLM systems via planning-phase prompt injection, revealing vulnerabilities in popular models like GPT-5 and Llama-3.3-70B. The study highlights the importance of heterogeneous model diversity for security. PlanFlip's four attacks can corrupt downstream sub-tasks, evading keyword filters and compromising system integrity.

A new arXiv paper by Alibaba researchers details a ReAct-style agentic setup integrating Large Language Models with SageMath, a powerful Computer Algebra System. This novel approach demonstrates substantial performance gains across frontier LLMs in solving research-level mathematical problems, significantly narrowing the capability gap between open-weight and closed models and paving the way for automated conjecture discovery.