Language models are rapidly becoming embedded in the interaction fabric of our society — as assistants, tutors, collaborators, and autonomous agents across domains from healthcare to software engineering to creative work. Their versatility makes them uniquely powerful but also uniquely hard to secure: privacy and security risks arise not just from model internals but from the rich, open-ended ways people interact with and through these systems.
Humans sit at the center of these challenges — as victims of privacy violations and manipulation, as intentional or unintentional facilitators of risk, and as bystanders whose data is swept up without consent.
HAIPS@COLM aims to bring a human-centered, interdisciplinary lens to LM privacy and security, building on the 1st HAIPS Workshop held at ACM CCS 2025. We aim to foster a two-way exchange between the COLM community and researchers in HCI, usable security, and policy — grounding human-centered approaches in what LMs actually do, and informing model design with how people actually experience them.
Key Dates
- Submission Deadline
- Notification of Acceptance
- Camera-Ready Deadline
- Workshop
Invited Speakers
-
Cornell Tech -
Virginia Tech -
Carnegie Mellon University -
Google -
ETH Zurich
Schedule
| Time | Session | Details |
|---|---|---|
| 9:00–9:10 | Opening remarks | |
| 9:10–9:40 | Invited talk | Florian Tramèr (ETH Zurich) Remote talk |
| 9:40–10:10 | Invited talk | Helen Nissenbaum (Cornell Tech) |
| 10:10–10:35 | Coffee break + posters | |
| 10:35–11:05 | Invited talk | Ruoxi Jia (Virginia Tech) |
| 11:05–11:35 | Invited talk | Maarten Sap (Carnegie Mellon University) |
| 11:35–12:05 | Invited talk | Nina Taft (Google) |
| 12:05–13:25 | Lunch | |
| 13:25–14:10 | Panel discussion | Panelists: Helen Nissenbaum, Ruoxi Jia, Maarten Sap, Nina Taft; Moderator: TBD |
| 14:10–14:15 | Break | |
| 14:15–14:25 | Spotlight talk | PrivacySIM: Evaluating LLM Simulation of User Privacy Behavior |
| 14:25–14:35 | Spotlight talk | Natively Unlearnable Large Language Models |
| 14:35–14:45 | Spotlight talk | Your Agent, Their Asset: A Real-World Safety Analysis of OpenClaw |
| 14:45–14:55 | Spotlight talk | Risk vs. Reward: Supporting Practitioners' Understanding of LLM Privacy Risks |
| 14:55–15:05 | Spotlight talk | Loose Lips: Understanding Privacy Leakage in Agentic Dialogue |
| 15:05–15:15 | Spotlight talk | Act on Your Behalf, Beyond Your Authorization: Benchmarking LLM Agent Authorization Alignment |
| 15:15–15:25 | Spotlight talk | Auditing Prompt Injection Defenses Before They Reach Users: What Strong Adversaries Reveal |
| 15:25–16:15 | Poster session + coffee | |
| 16:15–17:15 | Unconference discussions | |
| 17:15–17:30 | Closing remarks |
Call for Papers
HAIPS@COLM invites work that studies, critiques, and builds privacy, security, and safety protections for language models and agentic AI systems as technologies used by real people. We welcome submissions from both the core language-modeling community and the HCI, usable security, privacy, and policy communities.
We especially encourage work that connects technical mechanisms with human needs, including systems, tools, infrastructure, empirical studies, benchmarks, design work, policy analysis, audits, and position papers. Human-centered work need not always include a user study; technical papers are welcome when they clearly articulate the people, workflows, communities, or deployment contexts their work is intended to benefit.
Topics of Interest
Topics of interest include, but are not limited to:
- Human expectations, consent, and control over data used in LM systems
- Privacy- and security-preserving LM infrastructure for protecting people and institutions in real-world deployment contexts
- User-facing protections that help people understand and prevent LM privacy and security risks
- Human oversight, delegation, and accountability in agentic, personalized, and tool-using LM systems
- Evaluation and benchmarking of LM and LM-system privacy, security, and safety through a human lens
- Human impacts of adversarial LM-enabled privacy and security threats
- User vulnerability, manipulation, deception, and dark patterns involving language models
- Effects of language models used as simulated users, evaluators, auditors, or research participants
- LM-powered tools that support privacy, security, safety, and human agency
- Developer and organizational practices, risks, and responsibilities in LM design and deployment
- Human-centered policy and governance for LM privacy, security, and safety
Our goal is to bring together researchers who identify emerging risks, understand their human and societal implications, and build practical protections that make language-model systems safer, more private, more secure, and more accountable in real-world use.
Submission Format
- Long papers: up to 9 pages, excluding references and appendices.
- Short papers: up to 4 pages, excluding references and appendices.
- Submissions may use the COLM, NeurIPS, or EMNLP format.
- Accepted papers are non-archival, leaving authors free to submit revised versions to other venues.
Review Process
Review will be double-blind on OpenReview, with at least two reviewers per paper. Conflicts of interest will be managed through OpenReview’s COI detection and self-declared conflicts.
Awards
We aim to establish Best Paper and Best Student Paper awards to highlight outstanding work, with particular attention to early-career researchers.
Submit your paper via OpenReview: https://openreview.net/group?id=colmweb.org/COLM/2026/Workshop/HAIPS
Organizers
-
Northeastern University -
University of Notre Dame -
Johns Hopkins University -
Carnegie Mellon University -
humans& / Carnegie Mellon University
