Introduction
Artificial intelligence is no longer a theoretical discussion within social work. It is entering frontline practice, and the governance frameworks needed to manage that entry are not keeping pace.
In 2026, Social Work England published research confirming what many practitioners have been quietly experiencing for months. Eighty-three percent of respondents believed AI had the potential to reduce administrative burden. Eighty-six percent reported receiving no specific preparation on AI use during their professional training. And running through both findings — the optimism and the gap — was a concern that neither statistic adequately captures: when AI contributes to professional documentation, who remains accountable for what it says?
That question is not a technical one. It is a regulatory one. And in social work, where documentation functions simultaneously as a safeguarding record, a legal document, evidence of professional reasoning, and a future accountability mechanism, leaving it unanswered is not a minor oversight. It is a structural risk.
The Record Is Not Just Text
Case notes travel. They are read by inspectors, cited in courts, examined in safeguarding reviews, scrutinised by coroners, and reviewed by families and multi-agency partners. Each time a record is read, the implicit assumption is that it represents what the practitioner observed, judged, and decided — that the words on the page are a faithful account of professional reasoning.
When AI systems contribute to that record, the assumption becomes unreliable. The Research in Practice survey commissioned by Social Work England found that some practitioners had used publicly available tools such as ChatGPT, Claude, or Microsoft Copilot to support case recording, often without employer direction or clear governance frameworks. In those cases, the question of who authored the record — and who is professionally responsible for its accuracy — becomes genuinely difficult to answer.
This is what Stephen Hall has described as the travelling account problem. His work on Digital Narrative Care explores what happens to professional narratives as they move through AI systems, institutional processes, and summarisation layers. The risk, as Hall argues, is not always outright fabrication. Sometimes the greater danger is transformation — an account that remains superficially coherent while gradually drifting from the practitioner's original meaning, nuance, or contextual judgement. The wording becomes fluent. The accountability becomes diffuse.
In social work, that diffusion is not merely an ethical concern. It is a regulatory one. A practitioner who signs off an AI-assisted record that misrepresents what they observed, or fails to reflect the uncertainty they actually held, is professionally accountable for a document whose provenance they cannot fully trace.
What the Research Reveals
The Open University literature review commissioned alongside the Research in Practice survey identified recurring themes across the emerging literature on AI in social work: ethical uncertainty, transparency deficits, data governance concerns, and the risk of professional judgement erosion. The literature does not suggest AI should be rejected. It consistently points toward the need for strong governance, human oversight, clear accountability structures, and auditability of AI contribution.
The research investigation reinforced this. Respondents were broadly positive about AI's potential for administrative relief, and significantly less confident about its capacity to support decision-making or identify risk and need — the functions where professional judgement matters most. This is a sensible distinction that the governance frameworks being developed need to reflect. The use cases that attract the most interest are documentation and administration. The concerns that attract most anxiety are about accountability and accuracy. A governance approach that treats these separately will be inadequate.
What is striking about the Social Work England findings is not that they reveal a new problem. It is that they confirm a problem the sector has known was coming and has not yet equipped itself to address. Eighty-six percent of recently qualified social workers received no preparation for AI use in practice. They are now in roles where AI tools are available, sometimes employer-sanctioned, sometimes not, and they are navigating governance questions that their training did not prepare them for.
The Principle That Matters: Structure Over Generation
At ReporticaAI, the governing principle for AI-assisted documentation in regulated sectors is what we call Structure Over Generation. The distinction is important enough to be worth stating precisely.
A generative model that independently creates professional accounts — drafting case notes from minimal input, producing assessments from data patterns, summarising interactions without practitioner authorship — introduces accountability complexity that current governance frameworks cannot adequately manage. The record may be accurate. It may be fluent. It may satisfy formal requirements. But its provenance is unclear, and provenance is what professional accountability depends on.
A structured documentation system that organises, formats, and clarifies practitioner-authored material presents a fundamentally different profile. The practitioner remains the source of the content. The AI provides the structure. The accountability chain remains intact.
This is the distinction that the PAIDS-S framework — the Professional AI Documentation Standards for social care and safeguarding — was developed to operationalise. The framework establishes that AI systems used in regulated documentation must preserve traceable provenance, maintain practitioner ownership of records, make AI contribution visible and auditable, and ensure that the distinction between generated and authored content is never obscured. These are not aspirational principles. They are technical requirements that should govern how AI tools are selected, implemented, and governed in social work settings.
Why Governance Cannot Wait
One of the consistent failures in technology adoption across regulated sectors is retrospective governance — systems are implemented first, accountability frameworks are introduced later, often only after harm or regulatory failure has made the need visible. Social work cannot follow that path. The stakes are too high and the documentation too consequential.
The Social Work England research signals that adoption pressures are accelerating faster than governance maturity. That asymmetry creates risk at exactly the point where risk is least manageable — at the beginning, when norms are being set and habits are being formed.
The organisations that handle this well will not necessarily be those using the most AI. They will be those capable of demonstrating, when asked, that their documentation is accountable, their AI use is auditable, and their practitioners are the authors of the professional records they sign. That is a governance question before it is a technology question.
Social Work England, Skills for Care, and the broader sector are right to be convening the conversation this month. The question the summit is asking — how can AI enhance rather than replace the human relationships and professional judgement at the heart of social work? — is the right question. The answer begins with governance infrastructure at the point of document creation, not just at the point of procurement.
The accountability gap is real. The tools to close it exist. What is needed now is the governance commitment to use them.