In sociotechnical systems analysis, a handoff is the point at which responsibility for performing a function is transferred between components of a system — from a human to an automated component, from automation to a human, between automated components, or between organizations. The concept was developed as an analytic lens by Deirdre Mulligan and Helen Nissenbaum in "The Concept of Handoff as a Model for Ethical Analysis and Design," which argues that the moment of transfer, rather than the components in isolation, is where a system's values and risks are reconfigured (Source: https://papers.ssrn.com/sol3/papers.cfm?abstract_id=3784839).
The lens
The handoff lens directs attention to how a function is transferred and to what is gained or lost in the transfer: which actor or component holds responsibility, what assumptions are embedded in the interface between them, and how accountability, discretion, and error are redistributed. Treating handoffs as the unit of analysis reframes the governing question of a system from "what are the model's risks?" to "where are the handoffs, and what changes at each?" The authors present it as a model for ethical analysis and design, intended to evaluate sociotechnical systems in ethical and political terms rather than purely technical ones (Source: https://papers.ssrn.com/sol3/papers.cfm?abstract_id=3784839; https://afog.berkeley.edu/publications).
Use in AI risk governance
The handoff lens is named in Sociotechnical AI Risk Governance as the analytic approach compatible with a sociotechnical, rather than model-centric, orientation to AI risk — the framing advanced by Mulligan, Nik Marda, and Victor Zhenyi Wang (Knight Columbia, March 2026), which treats handoffs as the site of much of the risk in a system. It connects to the Levels of Autonomy for AI Agents (Feng-McDonald-Zhang framework), in which each level structures a distinct human-role and handoff pattern; to the Principal-Agent Problem Applied to AI; and to Meaningful Human Review, where requirements for human oversight can be read as specifications for how an AI-to-human handoff must be designed (Source: https://papers.ssrn.com/sol3/papers.cfm?abstract_id=3784839).
Relationships
- depends-on: Sociotechnical AI Risk Governance — the governance framing that adopts the lens as an analytic primitive.
- related: Levels of Autonomy for AI Agents (Feng-McDonald-Zhang framework), Principal-Agent Problem Applied to AI, Meaningful Human Review.
- related: Deirdre K. Mulligan — co-originator of the concept.
Provenance note
Page created 2026-06-15 (gap-scan) from the SSRN landing page and UC Berkeley AFOG publication listing for the originating chapter. It resolves a dangling [[concepts/handoffs]] reference relied on by Sociotechnical AI Risk Governance and Deirdre K. Mulligan. A foundational ingest of the Mulligan–Nissenbaum chapter would deepen the page.