person

Niloofar Mireshghallah

Carnegie Mellon postdoc; LLM privacy

Carnegie Mellon postdoctoral researcher (Allen School / CMU SCS); published widely cited work on LLM memorization, privacy attacks, and contextual integrity.

current Postdoctoral Researcher, Carnegie Mellon University

homepage @niloofar_mire

Strategy positions

Near-term harms firstendorses

Documented present harms outweigh speculative existential narratives

Argues LLMs systematically violate users' contextual privacy norms in ways that current alignment work does not address; privacy is a first-order alignment problem.

Models leak the wrong information to the wrong audiences. Solving this requires teaching them norms about which audiences should hear which kinds of information, a contextual integrity problem.

§ paperCan LLMs Keep a Secret? Testing Privacy Implications of Language Models via Contextual Integrity Theory· arXiv / ICLR· 2023· faithful paraphrase

Closest strategy neighbours

by jaccard overlap

Other people whose strategy tags overlap with Niloofar Mireshghallah's. Overlap is on tag identity, not stance; opposites can show up if they reference the same tags.

Adam Grant
shared 1 · J=1.00
Wharton organizational psychologist
Anna Lauren Hoffmann
shared 1 · J=1.00
UW iSchool; 'data violence' theorist
Anton Korinek
shared 1 · J=1.00
UVA economist; AI and inequality
Brian Merchant
shared 1 · J=1.00
Tech writer; 'Blood in the Machine' author
Clay Shirky
shared 1 · J=1.00
NYU emeritus; 'Here Comes Everybody', 'Cognitive Surplus'
Cynthia Breazeal
shared 1 · J=1.00
MIT Media Lab; social robotics; founder Jibo

Record last updated 2026-04-25.