Ben

Ben Maltbie

Hi! I just finished my master's degree at MIT studying EECS and entrepreneurship. Before MIT, I worked as a software engineer at Amazon on truck/train routing algorithms and large scale optimization problems for their fleet across North America and Europe. I completed my undergrad at Georgetown University where I double majored in computer science and economics.

In the past year, I've been trying to transition into working as an AI safety researcher. My research interests are on societal impacts of AI, behavioral evals (particularly sycophancy and diverse personas), and black-box control. My experience so far is largely with multi-turn evals, training reward models, and red-teaming.

Research

* indicates equal contribution.

Intersectional Sycophancy: How Perceived User Demographics Shape False Validation in Large Language Models
Benjamin Maltbie, Shivam Raval
Preprint. 2026.
Out-of-Distribution Generalization of Risk Aversion in Language Models
Kristina Zhang*, Junior Chinomso Okoroafor*, Benjamin Maltbie*, Andrew Lin*, Abhitej Bokka*, Elliott Thornley
Preprint. 2026. Under Review, NeurIPS 2026 Main Conference.
Agentic Reinforcement Learning for Competitor-Aware Customer Retention
Benjamin Maltbie
Preprint. 2026.

Writing

Arguing that sycophancy, left unaddressed, could cascade into a catastrophic risk to humanity.

Experience

Work Education
2025 - 2026
MIT Schwarzman College of Computing
MIT Social and Ethical Responsibilities of Computing (SERC) Research Fellow
Training reward models to instill risk-aversion in open-source models
2025
Supervised Program for Alignment Research
Research Fellow
Investigating complex persona and sycophancy interactions
2025
Verizon
Master's Thesis
Agentic Reinforcement Learning for Competitor-Aware Customer Retention
2024 – 2026
MIT
M.S. & MBA Student
Electrical Engineering & Computer Science (EECS), Leaders for Global Operations (LGO)
2020 – 2024
Amazon
Software Engineer
Transportation Optimization (MMPO)
2019
Jane Street
Strategy and Product Intern
Previously called Business Development
2017 – 2019
Maxlab: Laboratory for Computational Cognitive Neuroscience
Research Assistant
Leveraging Prior Concept Learning Improves Generalization From Few Examples in Computational Models of Human Object Recognition
2016 – 2020
Georgetown University
B.S. & B.A. Student
Computer Science & Economics. George F. Baker Scholar