§ Accepted submissions

Position provocations

The position provocations, case studies, and works in progress accepted to the workshop. Camera-ready PDFs are hosted here as authors release them, with SSRN or arXiv links where authors have posted them; some papers are shared with workshop participants only.

Accepted provocations

№ 01 · position paper · PDF
Soohwan Lee · Kyungho Lee
UNIST

AI agents mediate not only individual contributions but also the process through which many contributions become a team position. This matters when an AI-generated account of a meeting reaches another team, a manager, or an organization. Such an account may be accurate sentence by sentence while obscuring how the group reached its position, whose contributions shaped it, and which differences remain unresolved.

№ 02 · position paper · PDF
Akshara Raje
Ascendion

Proposals for accountability in agent-mediated work—AI use statements, detection, watermarking, content provenance—presuppose that the parties who would rely on them want them. Whether that demand exists has not been measured in the markets where AI-mediated work is exchanged between strangers. Our position is that accountability mechanisms in such markets are selected on cost incidence rather than on design, and that any mechanism imposing a cost on the verifier will not be adopted there.

№ 03 · position paper · PDF
Jared Ren
University of Washington

In spring 2026, our four-person graduate research team spent a quarter interviewing professional visual artists about how they defend authorship when generative AI enters their work. The artists had clear rules about where AI was allowed. We had none, and by the end of the quarter a word none of us could account for had traveled from a coding memo into our final presentation, where our advisor asked why we had chosen it. Nobody could say. I argue that AI does not need to write anyone’s work to dissolve contribution; it only needs to be available in the background of each member’s individual workflow.

№ 04 · position paper · arXiv · PDF
Hengzhi Ye
Peking University

Communities often respond to potentially AI-assisted work by asking three questions: Was AI used? Was that use disclosed? Can hidden use be detected? These questions place AI use itself at the center of accountability while overlooking a deeper problem: unowned judgment. Evaluations, claims, decisions, and creative directions can be shaped by AI with no accountable human or institution prepared to stand behind them.

№ 05 · position paper · SSRN
Diego Gómez-Zará
University of Notre Dame

As AI systems increasingly synthesize team discussions and contributions, they can make a group appear to speak with one voice before its members have developed that voice together. We use contribution dissolution to describe this transformation: individual inputs are reorganized into a shared representation in ways that can obscure how those contributions differed, related to one another, and shaped the resulting position.

№ 06 · work in progress · SSRN · PDF
Hauke Sandhaus · Ruixiang Han · Wendy Ju
Cornell Tech

Increasingly widespread generative AI adoption is changing the nature of group design work. To understand how ownership and design intention shift, we interviewed eighteen students from a graduate interactive device design course where projects are done in teams, each walking through their own repository and chat logs. First, design intentions can have histories their authors cannot recover. Second, intention forms downstream of the tool. Third, the group record dissolves.

№ 07 · position paper · SSRN · PDF
Donghyun Sohn · Harin Lee
Northwestern University · Ewha Womans University

Developers increasingly use repository files such as CLAUDE.md, AGENTS.md, and .cursor/rules to guide AI coding agents before they produce code. Code review usually starts later, once a patch already exists: a reviewer sees the diff and maybe an "AI-assisted" note, but not the instructions that shaped the patch. We argue that accountability for AI-assisted work should start before the patch, not only after it.

№ 08 · position paper · PDF
Sujit Kumar Kamaraj · Vineet Pandey
University of Utah

Group projects are increasingly shaped by AI tools that let students revise and reframe each other’s work invisibly. One solution to this problem is building better provenance tracking tools that record students’ work explicitly for instructors to see. However, such efforts fail to track contributions that happened outside the tracking environment. Yet these are often the moments where a student’s understanding, or lack of it, is formed, making their invisibility a pedagogical problem.

№ 09 · position paper · PDF
Anirban Mukhopadhyay · Kelechi Eze · Kurt Luther
Virginia Tech

As AI agents move from answering questions to observing teams, calling tools, and proposing next steps, it becomes difficult to say what the agent contributed, and how the team leveraged those contributions in their final result. Here, we present the OSINT Research Platform (ORP) as a design provocation for accountability infrastructure. My provocation is that accountability should be designed into the division of labor while work unfolds, rather than reconstructed from a final output or an AI-use log.

№ 10 · work in progress · PDF
Xiangzhe Yuan · Naimul Hoque · Lucas M. Silva
University of Iowa

Body doubling, the practice of leveraging another person’s presence to support focus, accountability, and sustained task engagement, has emerged as a promising productivity strategy, particularly for people with attention regulation difficulties such as ADHD. We investigate an AI body double system that combines EEG tracking for cognitive state estimation with proactive LLM support to provide adaptive co-regulation during reading, studying, and writing tasks. We conceptualize this influence as processual contribution, where AI supports how users carry out their work rather than directly producing task content.

№ 11 · position paper · camera-ready pending
Contribution Dissolution as the Bottleneck to Unleashing the Productivity Gains of AI
Yong Suk Lee
University of Notre Dame

The workshop frames contribution dissolution as a problem of attribution, originality, and accountability. I agree, but push further: contribution dissolution is also the binding constraint on organizational productivity in the AI era. Despite consistent experimental evidence of individual-level productivity gains from AI, organizational gains have not materialized. I argue that a key mechanism is the human verification bottleneck—generation scales, verification does not—and that this bottleneck persists precisely because we lack shared, explicit norms about what may be delegated to AI.

№ 12 · position paper · camera-ready pending
The Other Half of the Loop: Reception, Defensibility, and the Backstage of Agent-Mediated Work
Pia Tukkinen
Aalto University

This provocation accepts the workshop’s premise—that documented provenance is not witnessed contribution—and adds three things to it. First, agents have reached the receiving side of collaboration: a recipient can now perform the role of evaluating someone’s work without comprehending it, mirroring the dissolution the call locates in production. Second, the full record of how something was created with AI could in principle be produced, but producing it serves no one. Third, whether a contribution can be defended, and by whom, was distributed by role and seniority before agents arrived, and hierarchy often allocates defense away from the person who did the work.

№ 13 · case study · camera-ready pending
University Students’ Perceptions of Process Tracking for GenAI Attribution
Karina LaRubbio
Brown University

Generative artificial intelligence (GenAI) is changing the way that educators approach assessment and academic integrity by obscuring expectations for attribution, originality, and accountability. Some educators are using process tracking tools to monitor students’ behavior while they work for indicators of GenAI use. We conducted semi-structured interviews with eight university students to garner perspectives on opportunities and concerns presented by process tracking tools in post-secondary education. Participants described desires for more nuanced approaches to evaluating how rather than whether they used GenAI, though they questioned the ability of existing process tracking tools to accomplish this.

№ 14 · position paper · camera-ready pending
Before Accountability: Youth AI Use and Witnessable Contribution
Zao Li
Huazhong University of Science and Technology

This position provocation asks how AI-mediated work can become difficult to answer for once it enters a setting of evaluation, coordination, or accountability. Drawing on an exploratory discourse analysis of public discussions on Zhihu, a Chinese question-and-answer platform, I argue that young users often rely on generative AI to complete learning and career-related tasks, but the final artifact can reach a classroom, interview, or review forum with the formative process already obscured.

№ 15 · position paper · camera-ready pending
The Workshop Without Its Master: Context Provenance as the Neglected Layer in Agent-Mediated Work
Kai Lukoff
Santa Clara University

Researchers have proposed a growing set of instruments for recording contribution in AI-assisted work: AI use statements, edit logs, and fine-grained co-writing datasets. I argue, however, that they neglect the layer where authorship in agent-mediated work is increasingly decided: what the model knew about the author before the session began. I call the record of this layer context provenance.

№ 16 · case study · camera-ready pending
Verification Is the Hard Part: An Accountability Case Study of an Agent-Built DAW Bridge
Ziad Salah
Independent Researcher

Agentic coding tools can now reimplement existing systems, but the community has few published grounded accounts of how such reimplementations proceed and how their outputs are verified. We report on an extended field experience: the reimplementation of an open-source Model Context Protocol bridge that exposes the scripting API of FL Studio, a commercial digital audio workstation, to language-model agents. We argue that verification, not generation, is the hard part of reimplementation, and propose runtime-grounded verification, runtime surface discovery, and documented impossibility boundaries as candidate accountability norms.

№ 17 · position paper · camera-ready pending
Who Moved the Robot? Contribution Dissolution in Agentic Telepresence
Hyunju Kim
Cornell University

Agentic telepresence systems increasingly generate movements and gestures on behalf of remote users. This automation reduces operational burden, but it also makes it difficult to establish whether the remote user retains meaningful authority over the judgments a robot’s action expresses. Drawing on prior work and ongoing research in telepresence, I describe two forms of embodied contribution dissolution: temporal dissolution, in which control moves among actors over time, and structural dissolution, in which several actors jointly constitute one action from the outset.

№ 18 · position paper · participants only
Whose Style Is This? Visual Style as Witnessed Contribution in Agent-Mediated Design
Taeyoung Ko
UNIST

All accepted submissions are listed. PDFs hosted on this site are the authors’ camera-ready versions; where authors have also posted to SSRN or arXiv, those links are included. Entries awaiting a camera-ready version name the corresponding author and quote from the submitted abstract; full author lists follow with the final versions.

The collective output

The primary collective output of the workshop is the Accountability Infrastructure Casebook: a design provocation grounded in the concrete cases participants bring, outlining what collaborative systems and institutional norms would need to preserve for witnessed contribution to remain possible.

Accepted provocations are hosted on this site and, for additional visibility, posted to SSRN under an open license. Workshop participants will be invited to co-author a synthesis paper for CSCW or CHI.