MSc dissertation research / 01
Accountability for Agentic AI
Accountability for Agentic AI in Workplace Decision-Making: From Transparency to Enforceability
Research question
How can accountability for agentic AI in workplace decision-making move beyond transparency towards enforceable organisational arrangements?
Research design
The dissertation used reflexive thematic analysis of regulatory texts, industry standards and corporate governance policies spanning 2015–2025. The documentary corpus included the EU AI Act, NYC AEDT, ISO/IEC 42001 and IEEE 7000.
Analytical contribution
I operationalised accountability through two dimensions: answerability and enforceability. The analysis argued that records and documentation do not by themselves establish accountability where institutional authority to challenge decisions and impose sanctions is absent.
Framework and diagnostic tools
I developed an Interoperability Map connecting governance standards, design specifications and audit protocols. Two diagnostic tools—the Minimum Evidence Bundle for bias assurance and the Meaningful Human Control Test—focus attention on evidence, veto power and access rights.
Research development
The study provides a conceptual basis for further empirical investigation of how people exercise oversight and contest AI-mediated decisions inside organisations. Its application in organisational settings remains a question for further empirical research.