AI CV Screening: Is It Legal for a Small Business?

Updated 8 min read

AI CV screening is legal for a UK small business, provided a person genuinely considers every applicant or the firm applies the automated decision safeguards. The ICO’s 2026 recruitment report treats a tool that rejects below a pass mark as an automated decision. So is ranking applicants and reading only the top. Equality law applies either way, and the employer carries it.

AI CV screening is a software step that reads job applications, compares them with the role and then scores, ranks or rejects the applicants.

AI CV screening is legal in the UK because no law bans it, and the rules on automated decisions were relaxed in 2026. The Data (Use and Access) Act 2025 replaced the old Article 22 of UK GDPR, in force from 5 February 2026.

The ICO says the Act makes it easier to use automation to decide without a person involved. Two sets of rules still apply.

  • Data protection. Article 22A calls a decision solely automated when there is “no meaningful human involvement”. The ICO treats an automated rejection for a job as a significant decision, which brings the safeguards in.
  • Equality. Section 39 of the Equality Act 2010 bars discrimination in the arrangements an employer makes for deciding who gets a job.

Article 22B restricts solely automated decisions that use special category data, such as health, ethnicity or religion, to narrow conditions. CVs often give those details away.

The ICO’s guidance on automated decisions was still marked as draft as of November 2026, after a consultation that closed on 29 May 2026.

When does a person’s check count as a real decision?

A person’s check counts as a real decision when that person can change the outcome and has the authority and competence to do so. That is the test in Recruitment Rewired, the ICO’s 2026 report.

The report drew on over 30 employers that spoke to the ICO voluntarily between March 2025 and January 2026. Most believed their tools only supported decisions, yet the ICO frequently saw evidence that managers were unlikely to review lower-scoring candidates.

Its case study uses red, amber and green bands. The manager had been trained to review everyone, but prioritised green and only glanced at red. The ICO called that rubber-stamping and treated the red rejections as solely automated.

The same page says meaningful involvement must apply to every candidate, including those who score poorly.

Which AI CV screening set-ups count as automated decisions?

AI CV screening counts as an automated decision when the applicants the tool scores low are rejected without a person really considering them. The table applies the ICO’s published use cases to the set-ups a small firm is likely to buy.

The ICO has no use case labelled CV screening. Its closest is AI assessment of written applications, so the last two rows are Bedrock AI’s reading of its test.

Set-up Status What the small firm must do
Yes/no knock-out questions a person sets on essential requirements, such as right to work ICO: not an automated decision Keep each knock-out question to a genuine job requirement
Tool rejects applicants below a pass mark ICO: automated decision Apply the Article 22C safeguards and complete a DPIA
Tool ranks applicants, owner interviews the top few and rejects the rest unread ICO: automated decision Apply the safeguards, or read every application
Red, amber and green bands, with red applicants only glanced at ICO: solely automated, rubber-stamped Apply the safeguards, or give red applicants a proper read
Owner reads every application, with AI scores or notes alongside Bedrock AI’s reading: likely outside the rules if the owner can and does overrule Record why applicants were rejected, in the owner’s own words
Owner pastes CVs into a chatbot and takes its top five Bedrock AI’s reading: same as ranking, plus a data problem Stop, or read every CV and use a business account

Account settings are covered in is ChatGPT safe for business use.

What must you do if a tool makes the decision?

A firm letting a tool decide must give applicants the four Article 22C safeguards and say plainly that automation is used. The ICO found employers did not always apply them in full, often because they believed a person’s involvement took them outside the rules.

  1. Give information about the decision. The ICO stresses information specific to each person about how the decision was made.
  2. Let applicants make representations. A reply address that someone reads is the minimum.
  3. Provide human intervention. One employer offered this right with no method of handling requests, and the ICO expects a clear process.
  4. Let applicants contest the outcome. Decide who reviews it and how quickly.

Two privacy notice shortcuts failed the ICO’s review. A generic line saying screening “may be automated” was judged insufficient. Pointing applicants to the supplier’s privacy policy was poor practice, since the ICO says accountability lies with the employer using the tool.

The ICO also says automated recruitment is likely to require a DPIA. Our AI risk assessment template shows how to spot that, and the recruitment agencies page lists the ICO’s six buying questions.

Who is liable if the tool discriminates?

The employer is liable if an AI CV screening tool discriminates, because the Equality Act duty sits with the person deciding who to hire. The ICO’s report says organisations are responsible for the algorithms and tools they use, even when they did not build them.

Section 19 covers a neutral rule that puts people with a protected characteristic at a particular disadvantage. It is unlawful unless it is a proportionate way of achieving a legitimate aim. A filter on career gaps could disadvantage parents or disabled applicants.

A University of Washington study ran over 550 real CVs through three language models, and white-associated names were favoured 85% of the time. Those were US names and research models, so treat it as a warning.

Injury-to-feelings awards alone run from £1,300 to £12,600 in the lower band for claims presented in England and Wales from 6 April 2026. Employers also owe reasonable adjustments to job applicants, so a disabled applicant needs a route around the tool.

The ICO expects employers to ask developers about their bias testing and to run their own trials. Some employers it spoke to ran monthly bias reviews, which would mean little with a few dozen applicants.

Bedrock AI’s suggestion at that size is a swap test. Take ten CVs, change only the name or the dates that imply age, and see whether scores move. Ten CVs cannot prove a tool fair, but can catch one that is clearly unfair.

Is AI CV screening worth it for a small business?

AI CV screening rarely pays for a firm hiring a few people a year, because the time saved is worth less than the tool. Here is the sum for a 12-person firm, assumptions marked.

  • Applications: 4 roles a year × 40 applications each (assumed) = 160 applications.
  • Reading time: 160 applications × 3 minutes each (assumed) = 8 hours a year.
  • Cost of reading: 8 hours × £30 an hour (assumed loaded cost) = £240 a year.
  • Tool: Manatal’s entry plan, one example with AI scoring, lists $19 a user a month as of November 2026. One user at about £14 × 12 = £168 a year.
  • Saving: if the tool halves the reading (assumed), the firm saves £120 a year and is £48 worse off.
  • Break-even: £168 ÷ £30 = 5.6 hours saved. At 1.5 minutes saved per application (half of the assumed 3), that is about 225 applications a year.

The sum leaves out the DPIA, privacy notice and review hours. Those, or a dearer tool, push the break-even higher.

Below about 225 applications a year on these assumptions, read them yourself. Above that, a tool can pay if a person still considers each applicant or the safeguards are in place.

Bedrock AI’s view is that a small firm’s hiring problem is usually a weak job spec, and scoring 160 applications faster does not fix it. Diagnose where the hours go first. Our ROI calculator tests the sum against your own figures.

What is a safer way to use AI in hiring?

The safer way to use AI in hiring is to apply it before applications arrive and leave the reading to a person.

  1. Turn the job spec into criteria. Ask AI to list five must-haves and five nice-to-haves, then remove anything the job does not need.
  2. Set knock-out questions. Use yes/no questions on your application form for genuine requirements.
  3. Score applicants against the criteria yourself. A fixed list gives you a consistent read and a written reason for each rejection.

If you still want a tool, treat it as a tier 2 purchase and get the answers in writing. This guide is general information and not legal advice, so take advice on a live dispute.

FAQ

Do I have to tell applicants I use AI to screen CVs? Yes, applicants must be told how their information is used, and the ICO expects specific wording about automated screening. A generic line saying screening “may be automated” was judged insufficient in its 2026 report.

Can I paste CVs into ChatGPT to shortlist applicants? Pasting CVs into ChatGPT puts applicants’ personal data into a third-party tool, so a personal account is the wrong place for it. Taking the chatbot’s top picks and rejecting the rest unread also looks like the ranking set-up the ICO treats as automated.

Do I need applicants’ consent for AI CV screening? Consent is usually the wrong basis for AI CV screening, because the ICO doubts applicants feel free to refuse. Its 2026 report says legitimate interests is likely to be the most appropriate basis where no special category data is used. It expects a written assessment.

Does the EU AI Act apply to AI CV screening? The EU AI Act lists systems that filter job applications as high-risk in Annex III. A UK firm hiring only in the UK is outside its main reach. A UK firm hiring people into EU countries should take advice first.

How long can I keep CVs from rejected applicants? The ICO’s guidance says CVs from rejected applicants should not outlast the period for bringing a recruitment claim, without a clear business reason. Its 2024 review found some AI recruitment tools keeping data indefinitely, so check the supplier’s retention period and state it in your privacy notice.


Bedrock AI finds where AI will actually save your business time and money, before you spend anything on building. Book a free 15-minute call

Keep reading

More for owners

All articles
AI procurement AI Vendor Due Diligence Checklist for a Small Business An AI vendor due diligence checklist for owners: three tiers, so a note-taking app gets three questions and a tool wired into your CRM gets twelve. / 8 min AI risk assessment AI Risk Assessment Template for a Small Business An AI risk assessment template for a small firm: five questions, a 1-3 score and a clear approval rule, with three use cases scored in full. / 8 min AI policy AI Acceptable Use Policy: What a Small Firm Needs An AI acceptable use policy for a small business fits on one page: approved tools, a data traffic light, a human check and eight clauses you can copy. / 7 min
Book a free call