The Human Liability Sink: How Human Oversight Can Legitimate Automated Decisions While Laundering Responsibility
The statement that a human remains “in the loop” can describe either responsible control or an architecture for relocating responsibility. This paper develops the latter possibility. A human liability sink exists when a human ratification role helps legitimate automated reliance before harm and supplies a focal bearer of responsibility after harm, although substantive authority over the agenda, recommendation, or conditions of review remains upstream. The paper distinguishes decisional authorship, institutional attribution, blame, answerability, and remedy; proposes a Ratification Principle based on epistemic access, practical control, temporal opportunity, and protected refusal; and adds a Principle of Non-Consolidation under which final approval does not transfer authorship of upstream contributions. A finite conceptual model enumerates all assignments of fifteen institutional features to test the account’s logical relations with moral crumple zones, responsibility gaps, rubber-stamping, meaningful human control, and operator contestability. The model verifies consistency among stipulated predicates, not prevalence or moral truth. The analysis concludes with two operational diagnostics: the legitimacy counterfactual and the costly-dissent test.This is a conceptual preprint. The accompanying reproducibility package contains the standard-library Python model, specification, schema, expected results, complete enumerated outputs, deterministic tests, and release hashes. The enumeration tests consequences of stipulated predicates; it does not measure institutional prevalence, establish legal liability, or validate the predicates as empirical instruments.AI-assistance disclosure: The author originated the motivating question. AI assistance was used to develop and challenge conceptual distinctions, conduct bounded literature searches, construct and test the finite model, draft and revise prose, and run citation and consistency checks. The author determined the publication direction and accepts responsibility for the claims and final text.
Authors
- Ignacio Adrián LERER (ORCID: https://orcid.org/0009-0007-6378-9749)
Publication Details
- Journal
- Zenodo (CERN European Organization for Nuclear Research)
- Published
- 2026-09-05
- DOI
- https://doi.org/10.5281/zenodo.22354838
- Primary Topic
- Ethics and Social Impacts of AI
- Type
- preprint