Automated Decisions After the Data (Use and Access) Act: What UK Businesses Using AI Need to Know

The Data (Use and Access) Act 2025 rewrote the UK GDPR rules on automated decision-making from 5 February 2026. A practical guide for businesses building AI into decisions about people.

Dev Team Solutions 6 min read
Automated Decisions After the Data (Use and Access) Act: What UK Businesses Using AI Need to Know

If your business uses software to make decisions about people, the rules changed this year. On 5 February 2026 the automated decision-making provisions of the Data (Use and Access) Act 2025 came into force, replacing Article 22 of the UK GDPR with four new articles, 22A to 22D.

The change is often described as a relaxation, and in some ways it is. But it comes with firm safeguards, and as more businesses put AI into screening, scoring and approval processes, getting those safeguards right is now a design question for anyone building software. This guide is written from a developer’s point of view for business owners and product teams. It isn’t legal advice, and for anything high-stakes you should speak to a data protection specialist.

What Changed

Under the old Article 22, people had a general right not to be subject to a decision based solely on automated processing that had legal or similarly significant effects, with a short list of exceptions.

The new framework works differently:

ArticleWhat it covers
22ADefinitions. A decision is “based solely on automated processing” if there is no meaningful human involvement. A “significant decision” is one with a legal or similarly significant effect on someone.
22BRestrictions. Solely automated significant decisions using special category data, such as health or ethnicity, are only allowed in narrow circumstances, such as with explicit consent.
22CSafeguards. Where a solely automated significant decision is made, the organisation must put safeguards in place.
22DPowers for the government to say more, by regulations, about meaningful human involvement and significant effects.

In practice, for most personal data, solely automated significant decisions are now permitted, as long as the Article 22C safeguards are in place. For special category data, the old restrictive approach largely remains.

The Four Safeguards

Article 22C says the safeguards must include measures that:

  1. tell people about decisions made about them in this way
  2. let them make representations about those decisions
  3. let them obtain human intervention from the organisation
  4. let them contest the decision

These are not paperwork. Each one needs something in your software and your processes: a notice at the right moment, a way to respond, a person with the authority and information to look again, and a route to change the outcome.

Person reviewing and signing a document at a desk

”Meaningful Human Involvement” Is the Key Phrase

Whether a decision counts as solely automated depends on whether a human is meaningfully involved. The Information Commissioner’s Office (ICO) has been consulting on updated guidance, and its current draft is blunt: human intervention “cannot be tokenistic”, and spot checking an automated process doesn’t meet the requirement.

For the safeguards themselves, the ICO’s draft guidance says the person reviewing a decision should have access to the relevant data the system used, and the discretion and authority to change the outcome. A reviewer who simply clicks “approve” on whatever the system suggests is not meaningful involvement.

This matters for AI in particular. Large language models are very good at producing confident, plausible recommendations, which makes it easy for human review to drift into rubber-stamping. If you are relying on human involvement to keep a decision outside the solely automated rules, you need to be able to show that the human genuinely weighs the evidence.

Is Your Use Case Affected?

The rules only apply to decisions about people that have a legal or similarly significant effect. Some examples to help you think it through:

ExampleLikely to be a significant decision?
Automatically rejecting job applicants based on a CV scoreYes, it affects access to work
Setting credit limits or payment terms for individual customersLikely, it affects access to finance
Refusing a rental application based on an automated checkLikely, it affects access to housing
A chatbot answering general questions about your servicesGenerally no, it isn’t a decision about the person
AI drafting a reply that a member of staff reviews and sendsGenerally no, if the human genuinely decides
Forecasting stock levels or routing deliveriesNo personal decision involved

The grey areas are where most of the work is. A system that ranks applicants and only shows the top ten to a recruiter may be making a significant decision about everyone it filters out, even though a human makes the final choice.

Building It In: A Developer’s Checklist

When we build AI into decision-making software, these are the things we design in from the start:

  • Map the decisions. List every point where the system decides something about a person, and classify each one: significant or not, solely automated or not, special category data or not.
  • Decide where the human sits. For each significant decision, choose deliberately between a genuinely human decision supported by AI, or a solely automated decision with the Article 22C safeguards.
  • Give reviewers the full picture. Show the inputs, the factors that drove the recommendation and the relevant source data, not just a score.
  • Keep an audit trail. Record what the system recommended, what it was based on, who reviewed it and what they decided. If someone contests a decision months later, you need to reconstruct it.
  • Build the routes for challenge. Notices, a way to respond, and a workflow that puts contested decisions in front of someone with authority to change them.
  • Treat special category data separately. Keep health, ethnicity and similar data out of automated significant decisions unless one of the narrow conditions clearly applies.
  • Test for consistency and bias. Run the same cases through repeatedly, check outcomes across different groups, and keep the results.
Close-up of a white robotic hand

What’s Coming Next

The ICO’s guidance on automated decision-making is still marked as draft following its consultation earlier this year, and the Act also requires the ICO to prepare a statutory code of practice on AI and automated decision-making. Expect more detail on what good looks like, particularly around meaningful human involvement.

If you also sell into Europe, remember that the EU’s own rules sit alongside the UK’s. Our EU AI Act checklist for UK suppliers covers that side.

How We Can Help

We build AI systems and bespoke software for UK organisations, with human oversight, audit logging and clear decision records designed in from day one rather than bolted on after launch. If you are planning to use AI in decisions about customers, applicants or staff, or you want a second opinion on a system you already run, talk to us.

ai uk-gdpr data-protection automated-decision-making compliance
Share:

Let's Work Together

Get in touch today to discuss your project requirements.