When Are LLM Inferences Acceptable? User Reactions and Control Preferences for Inferred Personal Information Conference: Preprint; ; CCS: Security and privacy Usability in security and privacy CCS: Human-centered computing Empirical studies in HCI CCS: Security and privacy Social aspects of security and privacy
Kyzyl Monteiro, Minjung Park, Alexander Ioffrida, Angelina Sanna, Hao-Ping (Hank) Lee, Niloofar Mireshghallah, Yang Wang, Sauvik Das · 2026
Read When Are LLM Inferences Acceptable? User Reactions and Control Preferences for Inferred Personal Information Conference: Preprint; ; CCS: Security and privacy Usability in security and privacy CCS: Human-centered computing Empirical studies in HCI CCS: Security and privacy Social aspects of security and privacy by Kyzyl Monteiro, Minjung Park, Alexander Ioffrida, Angelina Sanna, Hao-Ping (Hank) Lee, Niloofar Mireshghallah, Yang Wang, Sauvik Das (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2605-10013.txt.
Description
Monteiro and seven co-authors study a privacy question most work skips: not whether language models can infer personal facts, but how users feel about those inferences and what controls they want. Asking about holidays can reveal income, and asking about medicine can reveal medical history, yet such inferences may expose more than users meant to share. The team built Reflective Layer, a tool that surfaces unstated inferences from users' own ChatGPT histories.
In a mixed-methods study, eighteen regular ChatGPT users judged 215 inferences drawn from their own chats. Against expectations, questions and interest beat worry and concern. Unease centred where guesses felt off the mark or mismatched to expected use. Users were also clearly less comfortable letting advertisers and third-party apps use inferences than platform providers, so acceptability turns on context and transmission, not content alone.
Catalog record
| CREATOR | Kyzyl Monteiro, Minjung Park, Alexander Ioffrida, Angelina Sanna, Hao-Ping (Hank) Lee, Niloofar Mireshghallah, Yang Wang, Sauvik Das |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 89 KB |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2605.10013