Your AI screens candidates. The law asks whether a person really decides.

UK GDPR Articles 22A–22D came into force on 5 February 2026. In March, the ICO reported that employers who believed they were using AI as decision support were, in practice, often making solely automated decisions — and wrote to 16 of them.

Express your interest

Six questions about how you hire and what you use. They tell us whether these obligations reach you, and what putting it right would involve.

Express your interest →

No card details. No obligation. Stephanie reads every one herself — and will say so if we are not the right fit. See our Privacy Notice.

Why this expertise matters now

Articles 22A–22D — inserted by the Data (Use and Access) Act 2025 — set out when automated decision-making in hiring is lawful, and what meaningful human review actually has to look like. The distinction that matters is not whether a person is involved. It is whether that person sees enough, and has enough authority, for the decision to be theirs.

5 Feb 2026
UK GDPR Articles 22A–22D in force (SI 2026/82), setting out when automated decisions in hiring are lawful
16
Organisations the ICO wrote to in March 2026, having found they were likely making automated hiring decisions without adequate safeguards
~300
Recommendations the ICO issued to AI recruitment tool providers following its 2024 audits

In its Recruitment Rewired report of 31 March 2026, the ICO reported that employers it engaged with often believed they were using AI only as decision support, while in practice decisions were being made with no meaningful human involvement. It wrote to 16 organisations it considered likely to be making automated decisions without adequate safeguards, obtained commitments that they would act, and said it will not hesitate to take further action if necessary. Separately, its 2024 audits of AI recruitment tool providers produced close to 300 recommendations, including on tools that allowed filtering by protected characteristics and inferred ethnicity from candidate names.

To be clear about where this stands: no UK recruitment agency has been publicly fined for automated decision-making in hiring. What exists is a regulator that has set out its expectations, told named organisations it thinks they fall short, and said what happens next. The gap between being told and being enforced against is the window in which records get put in order.

Two further points that catch agencies out. Since 19 June 2026, the DUAA 2025 has required every UK controller to operate a structured complaints process — a route a rejected candidate can use to challenge a decision they do not understand. And under the Equality Act 2010, liability for a discriminatory outcome rests with the employer who used the tool, not only the supplier who built it. “Our vendor handles compliance” is not a defence available to you.

AI, Recruitment & Regulation

View all insights →