AOP in 2026, why this still matters when the machine is helping
3-minute read. Meet Edgar and Delroy. Sit with one question.
Two men, identical needs
Imagine two men in their early fifties. Both have type-2 diabetes, poorly managed since a small stroke last year. Both are showing early signs of cognitive decline. Both have refused services in a previous assessment, and a hospital admission has triggered a re-referral. Both have a daughter doing more than she can sustain.
One is called Eddie Novak. He came from Krakow in 1991. His daughter is Kasia.
The other is called Delroy Campbell. He came from Jamaica as a child. His daughter is Shanice.
Same medical history. Same family pattern. Same risk picture.
They would not. Section 2 shows you the difference and the mechanism behind it. The rest of the module builds from that point.
What anti-oppressive practice is, in one paragraph
AOP is not a policy you tick. It is the recognition that social work is never neutral, that every assessment happens inside power relationships, and that some of the people we support face structural disadvantages the people designing systems around them cannot see. Two names shaped how the profession in England talks about it now: Lena Dominelli (anti-oppressive theory and practice) and Neil Thompson (the PCS analysis: Personal, Cultural, Structural).
The British Association of Social Workers Code of Ethics requires you to "challenge discrimination and oppressive policies and practices." The verb is challenge, which is active and ongoing. It assumes you will see things to challenge, regularly, in your daily work.
What changed in 2026
Generative AI now drafts paragraphs that look like a social worker wrote them. Fluent, fast, useful for overworked practitioners. And trained on a corpus that contains every bias the English-speaking world has produced for the last two centuries.
That is the problem this module exists to address, by giving you a structured habit, called VERA:H, that you bring to every AI-assisted record before you sign it. The aim is not to refuse AI, but to use it with the discipline the work needs.
Where this module is going
You are reading Section 1 of six. From here:
- Section 2 shows you exactly how AI encodes bias, with Edgar and Delroy as the worked example. The mechanism is statistical, not malicious, and once you see it you cannot unsee it.
- Section 3 teaches the VERA:H framework: Voice, Evidence, Reasoning, Attribution, Human. Five elements of a structured prompt that narrows the drift.
- Sections 4 and 5 drill into cultural specificity, deficit framing, and the protected characteristics beyond race where the same drift applies.
- And then the knowledge check to lock it in.
The H in VERA:H is what holds it all together. You are the human in the room, and the human at panel and in court when the case demands it. AI will not be there. You will.
Before you click into Section 2: think of one assessment you wrote in the last month. Which protected characteristics did the person hold? Hold that case in your mind as we walk through how AI would have drafted the opening paragraph.