The Implementation Gap: A Socio-Technical Analysis of AI-Assisted Redaction Adoption in UK Public Authorities

Document redaction is a critical but increasingly strained function within UK public authorities, where Freedom of Information (FOI) obligations coexist with stringent data protection duties. While AI is frequently positioned as a remedy to manual redaction's time and error risks, limited empirical evidence exists on whether public authorities are actually adopting such tools. This study uses FOI requests as an observational method to examine redaction governance and AI-readiness across 44 UK public authorities spanning healthcare, central government, and research-intensive higher education. We received 30 responses (68.2%) and analysed them using a mixed-methods approach. The findings reveal a pronounced implementation gap between AI's technical promise and organisational reality. Only one authority reported active AI tool use, describing a workflow where software-generated redactions are reviewed by staff. Half of responses indicated relevant policies were "not held," and only six authorities reported documented training. Most organisations relied on external guidance rather than formal internal standards, with respondents highlighting resource constraints, inconsistent decision-making, and difficulties with complex document formats. We contribute an FOI-based baseline of current practices and reframe AI-assisted redaction as a socio-technical capability dependent on recordkeeping, standardised guidance, and trained human oversight, proposing a staged hybrid model for safe automation experimentation.