Research Engineer Scale AI

Yash Maurya

I work on frontier AI safety and evaluation. I think about what can go wrong when autonomous agents get access to our lives (yes, you probably gave yours too much access), especially when they start interacting with other agents. My focus is on AI control, monitoring, and multi-agent safety.

I completed my master’s in Privacy Engineering at Carnegie Mellon University. Previously, I worked at BNY, Samsung, and Dynamo AI.

CV
Yash Maurya

Recent news

Updates & milestones
  1. HarnessOpt-Bench accepted to Findings of EACL 2027.

  2. ROK-FORTRESS accepted to NeurIPS 2026. I’ll be presenting it in Sydney in December.

  3. "ROK-FORTRESS: Measuring the Effect of Geopolitical Transcreation for National Security and Public Safety" received the Best Paper Award at the TAIGR Workshop at ICML 2026.

Earlier updates (18)
  1. "LHAW: Controllable Underspecification for Long-Horizon Tasks" accepted to the Lifelong Agents: Learning, Aligning, Evolving Workshop at ICLR 2026.

  2. "MoReBench: Evaluating Procedural and Pluralistic Moral Reasoning in Language Models, More than Outcomes" accepted to ICLR 2026. project page

  3. Attended the AI Risk Summit 2025 at the Ritz-Carlton, Half Moon Bay and conducted a red-teaming workshop titled "AI Red Teaming Room".

  4. My paper titled "Making Privacy-Preserving AI Accessible: A Practitioner-Oriented Framework" was accepted to the SUPA 2025 workshop at SOUPS.

  5. Started working at Scale AI as a Research Engineer.

  6. My submission titled "When Privacy Guarantees Meet Pre-trained LLMs: A Case Study in Synthetic Data" got accepted to PEPR'25.

  7. Started a new role as a Privacy Researcher at Carnegie Mellon University

  8. Graduated from Carnegie Mellon University with a Master's in Privacy Engineering

  9. Wrote an article on auditing real-world differentially private synthetic data

  10. My position paper on unlearning benchmarks in LLMs got accepted at SatML.

  11. Started collaborating with Meta to study "How Information and Control Shape Data Sharing and AI Engagement"

  12. Joined the AI Hub at BNY(Bank of New York Mellon) as an AI Governance Intern.

  13. Presented my work "Through the Lens of LLMs:Unveiling Differential Privacy Challenges" at PEPR'24 in Santa Clara fully funded by CMU

  14. Received Academic Achievement Award from the International Association of Privacy Professionals - Carnegie Mellon University's 2023-2024 IAPP Westin Scholar.

  15. Received the Project Olympus Spark Grant by Swartz Center for Entrepreneurship for building a Language Learning agent to help non-native speakers improve their speaking skills

  16. My work 'Guardrail Baselines for Unlearning in LLMs' got accepted at Secure and Trustworthy Large Language Models(SeT LLM) workshop at ICLR 2024

  17. Completed the Certified Information Privacy Technologist (CIPT) credential from the IAPP - International Association of Privacy Professionals.

  18. Won Space Theme in HackCMU'23 with our Space-JEDI Project

Research

* Equal contribution

2026

SG-LegalClarify evaluation pipeline: assess whether facts are sufficient, answer or clarify, then grade with IRAC rubrics
Paper figureSource ↗

arXiv preprint arXiv:2611.19643 · 2026

SG-LegalClarify: A Large-Scale Benchmark for Legal Reasoning and Ambiguity Recognition in Singapore Law

Ziwei Gu, Michael S. Lee, Drew Rein, Yash Maurya, Vikneswaran Kumaran, Darshini Ramiah, Chenguang Wang, Udari Madhushani Sehwag

Tests legal reasoning and whether models recognize when to ask for missing facts. Across 1,100 lawyer-authored questions in 10 domains of Singapore law, evaluations of 19 LLMs reveal persistent gaps in reasoning and selective clarification, even with web search.

Trusted execution architecture for held-out harness optimization
Paper figureSource ↗

Findings of EACL 2027 · Accepted

HarnessOpt-Bench: Evaluating LLMs at Harness Optimization

Varun Ursekar, Apaar Shanker, Yash Maurya, Shehab Yasser, Vijay S. Kalmath, Veronica Chatrath, Yuan Xue

Benchmarks whether frontier LLMs can improve an agent's prompts, tools, memory, and orchestration under a fixed evaluation budget. Results show that optimizer models matter more than the coding harness they use, with gains varying widely by task and seed.

Insights Generator architecture with input data, orchestrator, scout agents, investigator agents, and an insights report
Paper figureSource ↗

arXiv preprint arXiv:2605.21347 · 2026

Insights Generator: Systematic Corpus-Level Trace Diagnostics for LLM Agents

Akshay Manglik, Apaar Shanker, Kaustubh Deshpande, Jason Qin, Yash Maurya, Veronica Chatrath, Vijay S. Kalmath, Levi Lentz, Yuan Xue

A multi-agent system that turns large corpora of LLM-agent traces into evidence-backed diagnostic insights. Experts using its reports improved scaffold performance by 30.4 percentage points over the unmodified baseline.

MoReBench moral dilemma, candidate reasoning, and contextualized rubric criteria
Paper figureSource ↗

International Conference on Learning Representations (ICLR 2026)

MoReBench: Evaluating Procedural and Pluralistic Moral Reasoning in Language Models, More than Outcomes

Yu Ying Chiu*, Michael S. Lee*, Rachel Calcott, Brandon Handoko, Paul de Font-Reaulx, Raphaël Millière, Paula Rodriguez, Chen Bo Calvin Zhang, Ziwen Han, Udari Madhushani Sehwag, Yash Maurya, Christina Q. Knight, Harry R. Lloyd, Florence Bacus, Conor Downey, Mantas Mazeika, Bing Liu, Yejin Choi, Mitchell L. Gordon, Sydney Levine

Evaluates model reasoning on morally ambiguous scenarios, where multiple conclusions can be defensible. Expert rubrics test whether models identify trade-offs, justify recommendations, and reason across normative ethics frameworks.

LHAW pipeline: segment extraction, candidate generation and filtering, and agent execution
Paper figureSource ↗

Lifelong Agents: Learning, Aligning, Evolving Workshop at ICLR 2026

LHAW: Controllable Underspecification for Long-Horizon Tasks

George Pu, Michael S. Lee, Udari Madhushani Sehwag, David J. Lee, Bryan Zhu, Yash Maurya, Mohit Raghavendra, Yuan Xue, Samuel Marc Denton

Builds controllably underspecified versions of long-horizon agent tasks and validates their ambiguity through actual agent execution. The 285-task release measures whether agents seek clarification when missing information is genuinely outcome-critical.

ROK-FORTRESS translation and transcreation pipeline for Korean-language safety evaluations
Paper figureSource ↗

NeurIPS 2026 · Accepted · TAIGR Workshop at ICML 2026 · Best Paper Award

ROK-FORTRESS: Measuring the Effect of Geopolitical Transcreation for National Security and Public Safety

Yash Maurya*, Michael S. Lee*, Drew Rein, Bert Herring, Jonathan Nguyen, Kyungho Song, Udari Madhushani Sehwag, Jiyeon Cho, Kaustubh Deshpande, Yeongkyun Jang, Jiyeon Joo, Minn Seok Choi, Evi Fuelle, Christina Q. Knight, Joseph Brandifino, Max Fenkell

A bilingual safety benchmark that separates language effects from geopolitical grounding through an English-Korean transcreation matrix. It exposes safety and over-refusal behaviors that translation-only evaluations miss.

2025

Illustrated overview of five ML pipeline phases connected to privacy threats, tools, and trade-offs
Conceptual overviewSource ↗

IEEE Symposium on Privacy Expectations (ISoPE) 2025 & SUPA 2025 Workshop on Societal & User-Centered Privacy in AI

Making Privacy-Preserving AI Accessible: A Practitioner-Oriented Framework

Yash Maurya, Ibrahim Mohamed Anis Chhaya, Hana Habib

Practitioner-oriented framework that organizes concrete ML privacy mitigations, tools, and design patterns across the ML lifecycle to help teams operationalize privacy-preserving AI in real-world deployments.

2024

Anatomy of an LLM unlearning benchmark, showing model data, unlearning, forget and retain queries, and evaluation
Paper figureSource ↗

IEEE Conference on Secure and Trustworthy Machine Learning (SaTML), 2025

Position: LLM Unlearning Benchmarks are Weak Measures of Progress

Pratiksha Thaker, Shengyuan Hu, Neil Kale, Yash Maurya, Zhiwei Steven Wu, Virginia Smith

Shows that modest, benign changes to common unlearning benchmarks can expose recoverable target information or substantially worse retention loss, making reported progress look more reliable than it is.

The paper’s simulator output showing differentially private aggregation of location categories around Carnegie Mellon University
Paper figureSource ↗

arXiv preprint arXiv:2405.03903 (Advisor: Professor Hana Habib)

Unified Locational Differential Privacy Framework

Yash Maurya*, Aman Priyanshu*, Suriya Ganesh*, Vy Tran*

A privacy framework for aggregating sensitive location-based data while protecting individual privacy through differential privacy mechanisms.

Comparison of unlearning and guardrail baseline evaluation results
Paper figureSource ↗

ICLR 2024 Workshop on Secure and Trustworthy Large Language Models

Guardrail baselines for unlearning in LLMs

Pratiksha Thaker, Yash Maurya, Shengyuan Hu, Zhiwei Steven Wu, Virginia Smith

Simple prompting and filtering baselines can match fine-tuning on common unlearning evaluations, motivating stronger metrics that distinguish inference-time safeguards from genuine parameter-level forgetting.

The UsersFirst design and analysis methodology
Paper figureSource ↗

Twentieth Symposium on Usable Privacy and Security (SOUPS 2024 Posters) & SOUPS 2024 Workshop on Privacy Threat Modeling (WPTM 2024)

UsersFirst: A User-Centric Privacy Threat Modeling Framework for Notice and Choice

Tian Wang*, Xinran Alexandra Li*, Miguel Rivera-Lanas*, Yash Maurya*, Hana Habib, Lorrie Faith Cranor, Norman Sadeh

UsersFirst: A user-centric framework for identifying and mitigating privacy notice and choice threats, extending beyond LINDDUN

Complete trajectory diagram contrasting normal clients, a malicious client, and a client with heterogeneous data
Paper figureSource ↗

Course Project for Federated and Collaborative Learning Course(10-719) at CMU (Advisor: Professor Virginia Smith)

Is it Worth Storing Historical Gradients?

Yash Maurya*, Joong Ho Choi*, Yingxin Liu*

Studies label-poisoning detection in heterogeneous federated learning. In the tested settings, current model weights generally match or outperform historical-weight methods, avoiding the storage cost of retaining long update histories.

2022

2021

Poster comparison of an input dog photograph with Score-CAM, SS-CAM, and IS-CAM attribution heatmaps
Poster previewSource ↗

Responsible Computer Vision (RCV) Workshop at CVPR 2021

Improved variants of Score-CAM via Smoothing and Integrating

Rakshit Naidu, Haofan Wang, Soumya Snigdha Kundu, Ankita Ghosh, Yash Maurya, Shamanth R Nayak K, Joy Michael

Introduces smoothing and integration variants of Score-CAM to produce more localized attribution maps for CNN predictions. The poster compares SS-CAM and IS-CAM with existing methods using visual examples and quantitative explanation metrics.

2020

Speaker bio

For event organizers

Yash Maurya is a Research Engineer at Scale AI, where he works on frontier AI safety. His research looks at how to keep increasingly capable AI agents in check, with a focus on AI control, monitoring, alignment auditing, and multi-agent safety. He came to safety through privacy, and his earlier work on LLM unlearning, red teaming, and differential privacy shaped his interest in how models behave under adversarial pressure. He earned his Master’s in Privacy Engineering from Carnegie Mellon University. Before Scale, he worked on AI governance and evaluation at BNY, federated learning at Dynamo AI, and recommendation systems at Samsung. His work has been published at leading machine learning and security conferences, including NeurIPS, ICLR, ICML, SaTML, and USENIX.

Notes & tools

Paper figure

Figure source