Anti-deniability design emerges independently in ethics and in doctoral supervision: Köbis et al. 2025 prescribe that 'delegation interfaces that make it easier for principals to claim ignorance of how the machine will interpret their instructions should be avoided,' while Gao & Zhang 2026's accountable reconstruction pushes students 'to connect AI outputs with source evidence, methodological assumptions, and argumentative responsibility.' Both force the human to retain visible ownership of machine-mediated outputs.
Köbis et al.'s cited text contains the verbatim prescription to avoid delegation interfaces that let principals 'claim ignorance of how the machine will interpret their instructions,' and Gao & Zhang's cited text contains the verbatim statement that students were pushed 'to connect AI outputs with source evidence, methodological assumptions, and argumentative responsibility,' supporting the claimed convergence on retaining visible human ownership.
Written by Kimi K3 via Ollama Cloud · checked by GLM-5.3 via Ollama Cloud · 15 Oct, 04:45
Source chain
AEvery quote below was checked, without a model, to appear verbatim in its source.
- 01
“delegation interfaces that make it easier for principals to claim ignorance of how the machine will interpret their instructions should be avoided”
Full-text passage · no page number · evidence E38
Consequently, ensuring that principals always have an option to not delegate, or making this option the default, could in itself curb the adverse effects of machine delegation. Most importantly, delegation interfaces that make it easier for principals to claim ignorance of how the machine will interpret their instructions should be avoided. In this regard, it may be helpful to better understand the moral emotions that principals experience when delegating to machines under different interfaces. We collected many measures of such moral emotions as exploratory exit questions but did not find any clear interpretation. We nevertheless report these measures for interested researchers in the Supplementary Information (the ‘moral emotions’ sections for each of the four studies) and Supplementary Fig. 1. Our protocols missed many of the complications of other real-world delegation possibilities. Die rolling and tax evasion have no social component, such as the possibility of collusion59–61. Future research will need to explore scenarios that involve collaboration within teams of machine and human agents, as well as their social history of interactions62–64.
Passage read from www.ebi.ac.uk, which may be a preprint rather than the published version.
- 02
“to connect AI outputs with source evidence, methodological assumptions, and argumentative responsibility”
Full-text passage · no page number · evidence E17
These studies indicate that doctoral writing is not merely a language-support task; it is a site where scholarly voice, methodological accountability, and disciplinary membership are formed. The mechanism developed here adds that GenAI-supported doctoral writing and research planning should be evaluated by whether AI-supported materials are reconstructed into claims, methods, and arguments that the doctoral student can explain and defend. Supervisory accountability was important because it made AI-supported reasoning visible. When supervisors asked students to mark AI-influenced parts and explain why suggestions were accepted or rejected, students were pushed to connect AI outputs with source evidence, methodological assumptions, and argumentative responsibility. This resonates with recent doctoral education research showing that GenAI can reshape supervisory and peer dynamics, create tensions between productivity-oriented research practices and traditional academic norms, and require clearer renegotiation of roles, responsibilities, and communication practices in doctoral training (Lai et al., 2025).
Passage read from www.ebi.ac.uk, which may be a preprint rather than the published version.