The Redundancy Map: HR Roles in the Age of Generative AI

From résumé screening to people analytics, HR sits at the intersection of the two forces this whole programme is about: AI can now do a great deal of the work, and European law is most watchful exactly where it does. Recruitment, worker management, and access to essential services are named high-risk in Annex III; people data is personal data at its most sensitive. The redundancy map is therefore also a compliance map.

The honest framing is not that AI replaces HR. It is that HR's task-mix is being resorted, some tasks automated, some amplified, some newly created, and the function's job is to manage that resorting for the whole organisation while getting its own house in order first.

Automated: the high-volume, structured tasks, interview scheduling, first-pass CV screening, routine reporting. Note the trap: CV screening is highly automatable and squarely high-risk, so automating it without bias-testing, human oversight, and documentation is the fastest route to an AI-Act problem. Amplified: coaching, development, sensitive employee-relations casework, organisational design, the human-judgment work that becomes more valuable as production gets cheap.

Created: the roles that make AI safe and effective in people processes, AI-governance and bias-audit specialists, worker-information-rights owners, and the designers of human-in-the-loop hiring. In Europe these are not overhead; they are the licence to use AI in HR at all.

Screening models, people analytics, and coaching assistants run on employee and candidate data that is personal, often special-category-adjacent, and politically charged. Running it on a third-country API multiplies the transfer, bias-audit, and worker-rights problems rather than solving them. Self-hosted, governed inference lets you show your work, to a DPO, a works council, and a regulator, which is the difference between using AI in HR and being allowed to.

The frontier model still has a place: non-personal tasks like drafting job-family frameworks or summarising public labour-market research. The bright line is the person. The moment a real candidate or employee is in the data, the default is sovereign.

Map every HR-AI use against the Annex III high-risk tests and your data-sensitivity scale; bias-test and document anything in screening or evaluation before it runs. Move the sensitive workloads to sovereign infrastructure your DPO can defend.

Then manage the resorting as a reskilling programme, not an HR event: fund paths from automated tasks into the amplified and newly-created roles. HR should model, for the whole company, the humane transition it is being asked to lead.