New research warns of 'contribution dissolution' as LLM agents blur attribution in teamwork

Researchers from multiple institutions (including Imteyaz, Rifat, Ramesh, Rick, Hosio, Sandhaus, Sarkar, Riedl, and Savage) have published a workshop paper on arXiv (ID 2607.26387) confronting a problem they call 'contribution dissolution.' In agent-mediated collaboration, LLM agents don't just facilitate—they actively synthesize, reformulate, and draft outputs within workflows. This undermines the social conditions needed for contributions to be witnessed, attributed, and held accountable. The problem begins at the individual level, where workers themselves feel uncertain about what is genuinely theirs, and propagates through teams, collapsing the trust that sustains productive intellectual exchange.

The authors critique current accountability measures—AI use statements, watermarking, provenance logs—as treating a documentation problem rather than addressing the structural conditions under which accountability is produced. Through position statements, mapping exercises, and hands-on activities, the workshop aims to surface deeper infrastructural challenges. The ultimate goal is a shared research agenda and foundations for an infrastructural response. This is particularly urgent as LLM agents become embedded in research, design, writing, and decision-making across industries, raising fundamental questions about authorship, credit, and responsibility in collaborative knowledge work.