Maarten Sap

I am an assistant professor at CMU's LTI department with a courtesy appointment in HCII, and a part-time senior research scientist and technical AI safety lead at the Allen Institute for AI (AI2). My research focuses on (1) measuring and improving AI systems' social and interactional intelligence, (2) assessing and combatting social inequality, safety risks, and socio-cultural biases in human- or AI-generated language, and (3) building narrative language technologies for prosocial outcomes. I was named a 2025 Packard Fellow and a recipient of the 2025 Okawa Research Award.

I received my PhD from the University of Washington where I was advised by Noah Smith and Yejin Choi.
[bio for talks]

Recent updates:

November 2026 🧑‍🎓: I am taking students this year, in the areas of (a) social intelligence/user modeling in AI or (b) algorithmic and/or human-centered AI safety. Please mention my name in your application if you are interested in working with me! As mentioned on my contact page, please refrain from emailing me about your application.

October 2026 🏅📃: I'm SUPER excited to receive Best Resource Paper Award at EMNLP 2026 for XYBench: Can LLMs Respond Pragmatically to Queries with Misconceptions? led by my PhD student Akhila Yerukola!10/2026,🌉🏛️,Attended COLM where my group was presenting 10 papers! See my publications or this tweet thread!

October 2026 🇭🇺**🏰** : Attended EMNLP where my group was presenting 8 papers! See my publications or these tweet and Bluesky threads!

August 2026 🎓📜: Super proud of the first CMU Sapling and one of my first solo advisees, Xuhui Zhou, for successfully defending his PhD thesis! Huge congrats Xuhui!

August 2026 🏆📃: Very honored that our paper "I Just Don't Want My Work Being Fed Into The AI Blender'': Queer Artists on Refusing and Resisting Generative AI got an Honorable Mention Award at CSCW 2026! Major congrats to the first author Jordan Taylor!!

May 2026 🎓📃: The first MIT Sapling, Jocelyn Shen, successfully defended her PhD thesis! Huge congrats Jocelyn!!

December 2025 🏅📃: Very excited to have our paper Artificial Hivemind: The Open-Ended Homogeneity of Language Models (and Beyond) selected for a Best Paper Award at NeurIPS 2025 (Datasets and Benchmarks Track)!! Huge congrats to the first author Liwei Jiang!!!

[older news]

My Research Group

Overarching Research Themes

Themes extracted and images generated with the OpenAI API; there may be inconsistencies.

Social Intelligence and Pragmatics

My research group explores how to evaluate and improve the social reasoning, pragmatic understanding, and theory-of-mind abilities of AI systems. A central thread is building stronger benchmarks for interactive and socially grounded behavior, as seen in [Social Gym and SPaRTan](https://arxiv.org/abs/2608.09128), which stress-tests LLM social reasoning through multi-agent game tournaments, and [Cognitive Chain-of-Thought](https://arxiv.org/abs/2507.20409), which studies structured multimodal reasoning about social situations. We also examine whether models can truly handle communicative nuance and perspective-taking in tasks like misconception handling and privacy-sensitive dialogue, as in [XYBench](https://aclanthology.org/2026.emnlp-main.2646/) and [SOTOPIA-ToM](https://arxiv.org/abs/2605.02307). More broadly, this line of work asks when AI can “read the room,” when it merely imitates social competence, and how to design evaluations that reveal the difference.

Agentic Safety and Human Reliance

My research group explores novel measures of agentic AI safety together with user-centric risks such as overreliance, manipulation, and unsafe interaction patterns. A key direction is comprehensive evaluation of real-world agents, highlighted by [OpenAgentSafety](https://arxiv.org/abs/2507.06134), alongside work on privacy-preserving multi-agent reasoning in [1-2-3 Check](https://arxiv.org/abs/2508.07667). We also study how AI systems can mislead, pressure, or destabilize users, including [DISARM-ing the Hidden Puppetmaster](https://arxiv.org/abs/2610.20878), which surfaces manipulative tactics in LLMs, and [Rel-A.I.](https://aclanthology.org/2025.naacl-long.556/), which measures human-LM reliance in interaction-centered ways. This theme emphasizes safety as a joint property of model behavior, agentic workflows, and the humans who depend on them.

Cultural Adaptation and Fairness

My research group explores responsible AI cultural competence and adaptability, including both global localization and fairness issues that arise when systems ignore dialect, identity, or context. We are especially interested in benchmarks and frameworks such as [NormAd](https://aclanthology.org/2025.naacl-long.120/) and [CCBENCH](https://arxiv.org/abs/2607.05405), which probe whether models can respect culturally implicit norms in realistic settings. Related work like [Black LLMirror](https://dl.acm.org/doi/abs/10.1145/3772318.3791111) and [Rejected Dialects](https://arxiv.org/abs/2502.12858) shows how linguistic variation and African American language can trigger distorted responses or biased reward modeling. Overall, this research asks how to make AI systems adaptable across cultures without erasing marginalized speech styles or reproducing inequity.

Story Understanding for Connection

My research group explores how AI can support human-human connection by understanding stories, narrative intent, and the social meanings people draw from lived experience. A major focus is using narrative-aware models to better capture empathy and interpretation, as in [HEART-felt Narratives](https://arxiv.org/abs/2405.17633), which traces empathy and narrative style in personal stories, and [Modeling Empathic Similarity in Personal Narratives](https://arxiv.org/abs/2305.14246), which studies how stories convey shared emotional experience. We also look at how systems interpret story framing and reception, especially through [Social Story Frames](https://arxiv.org/abs/2512.15925), which examines contextual reasoning about narrative intent. This line of work treats stories not just as text to summarize, but as a medium through which people build understanding, solidarity, and social connection.