10/23/2026
Congratulations to Fraol Batole whose paper: RefineAct: Automatic Runtime Verification of LLM Agent Actions has been awarded an ACM Distinguished Paper Award at ASE 2026!
Batole shares, "RefineAct tackles a problem that's becoming urgent as AI agents take on more autonomous tasks. Agents today operate without a real specification of what they should and shouldn't do, which means nothing checks their actions before they execute. Our framework automatically turns a user's natural-language instructions into a formal specification, then verifies each action the agent proposes against it in real time. This helps catch unsafe or misaligned behavior before it causes harm, rather than after. In our evaluation, it cut failure rates nearly in half while also improving how well agents completed their tasks.
I began this research after reading about cases where an AI deleted important data even though it was told not to. I wanted to find a way to make sure these assistants do what people actually ask, and nothing more, so people can trust them with real responsibility.
Winning this award means a lot, mainly because it's a sign that this problem (i.e., making AI agents safer and more trustworthy as they take on more responsibility) genuinely matters to the community, not just to me. I'm grateful to my advisor, Dr. Hridesh Rajan, and my co-author, Dr. Foutse Khomh, for their guidance throughout this project, and I'm looking forward to continuing to work on this line of research."
