Ali Iranmanesh

Senior CS @ Honors College, Penn State.

I work on AI safety and security.
My intrest is studying AI systems adversarially to find the failure modes that regular evaluations miss, understand the harms they can cause, and figure out what systems actually need to defend against.

email / scholar / LinkedIn / GitHub

Ali Iranmanesh

Publications

NeurIPS 2026

Oral

top 0.4%

HearSayBench: Can LLMs Navigate from Abstract Human Rights to Lived Lives?

Ali Iranmanesh*, Ava Iranmanesh*, Sobhan Lotfi*, Liwei Jiang

A benchmark probing whether models recognize the unstated structural constraints that make standard advice impossible for people in adversarial conditions. Across 14 frontier and open-weight models, we find a systemic 25% drop between understanding a situation and giving advice that respects it. Oral at NeurIPS 2026.

project page

arXiv

Preprint

Not What You Asked For: Characterizing Typographic Attack Surfaces and Harm in Embodied Vision-Language Agents

Ali Iranmanesh, Peng Liu

A characterization of typographic attack surfaces and harm in open-vocabulary embodied vision-language agents. We demonstrate how physical text perturbations can override a model's visual judgment and poison its persistent 3D semantic map, propagating through planning and manipulation to make the robot physically grasp and deliver the wrong object without modifying the underlying model.

arxiv

experience

2026

Jul to present, Los Angeles

Research Assistant, UCLA

Supervised by Yuan Tian

Investigating robustness of vision-language-action models: adversarial attacks against the binding stage, where the model decides which object in a scene an instruction refers to.

2025

Aug to present, State College

Research Assistant, Penn State University

Supervised by Peng Liu

AI security and robustness of embodied systems: typographic attacks against vision-language models in household manipulation robots. Also uncovering pointer-alignment bugs in Rust packages that crash programs in resource-constrained and embedded systems.

2025

Nov to Present, State College

Co-Founder, Perkie

An AI-native fintech app for optimizing credit card rewards and maximizing cashback.

2024

Summer, Hong Kong

Exchange Undergraduate Researcher, HKUST

Performance optimization of database management systems on GPUs, focused on advanced querying algorithms for accelerated workloads.

honors & awards

2026

B.&L. Gall Open Doors, Whitney Open Doors Honors, Chancellor's Scholarship

Full-cost support awarded to Schreyer Scholars for exceptional academic and leadership achievement.

2026

Erickson Discovery Grant

Awarded to support independent undergraduate research under faculty mentorship.

10 out of 89,000 students university-wide

2020

Gold Medal, Persian Young Physicists' Tournament

National team qualification phase for the International Young Physicists' Tournament.

service & teaching

2026

Reviewer, ICML 2026

Reviewed research submissions on LLM evaluation and AI safety for a top-tier machine learning venue.

2025

to present

Teaching Assistant, Penn State University

Head TA for Data Structures and Algorithms, Computer Organization and Architecture, and Communication Networks.