Consent and compliance

ICO guidance on AI in recruitment (2026): audit findings, automated decisions and what UK employers must do

On this page
  1. What has the ICO said about AI tools in recruitment?
  2. What did the ICO's audits of AI recruitment tools find?
  3. What should recruiters ask before buying an AI recruitment tool?
  4. What do the UK GDPR rules on automated decisions say now?
  5. When is a hiring decision "solely automated" in the ICO's view?
  6. What must employers tell candidates, and when?
  7. Which lawful basis does the ICO expect for AI in hiring?
  8. What does the ICO expect on bias monitoring?
  9. What should recruitment agencies check?
  10. Checklist for UK recruiters and agencies
  11. How does this compare with the EU and the US?
  12. Questions people ask

The UK Information Commissioner's Office (ICO) does not ban AI in recruitment, but it expects employers and recruitment agencies to treat AI hiring tools as high-risk processing under the UK GDPR: a DPIA before use, a documented lawful basis, clear information to candidates, bias monitoring with real data, and contracts that say who is controller. Its November 2024 audit of AI recruitment tool providers made 296 recommendations, and its March 2026 report "Recruitment rewired" found that many employers are likely making solely automated hiring decisions, which since 5 February 2026 must carry the safeguards in UK GDPR Articles 22A to 22D: information, the right to make representations, human intervention and a way to contest the decision. As of November 2026, the ICO's updated automated decision-making guidance and its recruitment and selection guidance are still drafts.

Not legal advice. This page summarizes the ICO's own reports and draft guidance, the UK GDPR as amended by the Data (Use and Access) Act 2025, and the commencement regulations, each linked, as of November 2026. Two of the ICO documents are drafts and will change. Confirm your position with your data protection officer or a qualified UK adviser.

What has the ICO said about AI tools in recruitment?

Five documents carry the ICO's position. Two are reports on what it found, one is the law it applies, and two are guidance still being finalized.

Rule or documentWho it coversWhat it requires or saysIn force from / statusSource
UK GDPR Articles 22A to 22D (automated decision-making), inserted by section 80 of the Data (Use and Access) Act 2025 Any controller taking a significant decision based solely on automated processing, including employers and agencies deciding who advances Safeguards: information about the decision, representations, human intervention, contest; tighter limits for special category data 5 February 2026 (decisions taken on or after that date) DUAA s. 80; SI 2026/82
AI tools in recruitment: audit outcomes report Developers and providers of AI sourcing, screening and selection tools, and the recruiters who buy them Seven key recommendations: fairness, transparency, data minimisation, DPIAs, controller and processor roles, processing instructions, lawful basis Published 6 November 2024 (audits August 2023 to May 2024) ICO report (PDF)
Key data protection questions for recruiters Organisations thinking of buying an AI recruitment tool Six questions to ask before procurement, starting with "Have you completed a DPIA?" Published 6 November 2024 ICO news
Recruitment rewired report Employers and recruitment platforms using automated decision-making in hiring Apply the Article 22C safeguards or build meaningful human involvement into every decision; improve transparency; monitor bias Published 31 March 2026; employers written to with deadlines ICO report
Draft guidance on automated decision-making and profiling (updated for the DUAA) All controllers How Articles 22A to 22D apply, lawful bases, special category data, safeguards Draft; consultation 31 March to 29 May 2026; no final date announced ICO draft guidance
Recruitment and selection guidance Employers and agencies recruiting in the UK Whole-process guidance from advertising to deletion Draft (consulted December 2023 to March 2024); listed as drafting, Winter 2026 ICO guidance plan

What did the ICO's audits of AI recruitment tools find?

Between August 2023 and May 2024 the ICO ran consensual audits of organisations that build or supply AI tools for sourcing (matching people from candidate databases), screening (scoring CVs, predicting interest or success) and selection (game-based assessments, scoring written answers and "text transcriptions of in-person or video interviews"). Biometric tools such as emotion detection and generative AI tools were outside the scope. The outcomes report records 296 recommendations and 42 advisory notes; 97% of recommendations were accepted and 3% partially accepted, and none was rejected. The findings recruiters should know:

  • Filtering by protected characteristics. Some tools had search features that let recruiters filter out candidates with certain protected characteristics.
  • Inferred demographics. Some providers estimated gender, ethnicity or other characteristics from a candidate's name or profile instead of asking. The ICO says inferred information is still special category data, was often processed without a lawful basis or the candidate's knowledge, and is not accurate enough to monitor bias.
  • Too much data. Some tools scraped and combined information from millions of people's job networking and social media profiles to build databases recruiters could market vacancies to, usually without the people knowing.
  • Wrong roles. Several providers called themselves processors when they were controllers, and some tried to pass all compliance responsibility to recruiters through vague contracts.

The report's DPIA findings (missing data flow maps, no assessment of necessity, risks to the organisation instead of to people) are worked through section by section in our DPIA template for AI note takers, so they are not repeated here.

The seven key recommendations

  1. Fairness: monitor for fairness, accuracy and bias issues and act on them; accuracy "better than random" is not enough where the AI decides without human intervention.
  2. Transparency and explainability: tell candidates what personal information the AI processes, the logic involved, and whether their data trains or tests the AI.
  3. Data minimisation and purpose limitation: collect the minimum needed and do not reuse it for an incompatible purpose.
  4. DPIA: complete one before processing and update it as the tool changes.
  5. Controller and processor roles: define them for each processing activity and write them into contracts and privacy information.
  6. Explicit processing instructions: where the provider is your processor, give written instructions on data fields, purposes, outputs and safeguards, and check it follows them.
  7. Lawful basis and additional condition: identify both before processing, and complete a legitimate interests assessment where you rely on legitimate interests.

What should recruiters ask before buying an AI recruitment tool?

The ICO's six questions, published with the audit report on 6 November 2024:

  1. Have you completed a DPIA?
  2. What is your lawful basis for processing personal information?
  3. Have you documented responsibilities and set clear processing instructions?
  4. Have you checked the provider has mitigated bias?
  5. Is the AI tool being used transparently?
  6. How will you limit unnecessary processing?

What do the UK GDPR rules on automated decisions say now?

The Data (Use and Access) Act 2025 replaced the old Article 22, which gave people a right not to be subject to solely automated decisions with only listed exceptions, with four new articles. Section 80 came into force on 5 February 2026 under the Commencement No. 6 Regulations (SI 2026/82), and the new rules apply only to decisions taken on or after that date.

  • Article 22A, definitions. A decision is based solely on automated processing "if there is no meaningful human involvement in the taking of the decision," and it is significant if it produces a legal or similarly significant effect. Rejecting a job application is the kind of decision the ICO treats as significant. When judging human involvement you must consider "the extent to which the decision is reached by means of profiling."
  • Article 22B, restrictions. A significant decision based entirely or partly on special category data (health, ethnicity and so on) may not be solely automated unless the candidate gave explicit consent, or the decision is necessary for a contract or required or authorised by law and the substantial public interest condition in Article 9(2)(g) applies. A solely automated significant decision also may not rely on the new "recognised legitimate interests" basis in Article 6(1)(ea).
  • Article 22C, safeguards. Where a solely automated significant decision is taken, the controller must have measures that give the person information about the decision, let them make representations, let them obtain human intervention, and let them contest it.
  • Article 22D, regulations. The Secretary of State may set by regulations what does or does not count as meaningful human involvement or a similarly significant effect, and add to the safeguards.

The practical shift for UK employers: a solely automated rejection is no longer presumptively prohibited, but it is only lawful with the safeguards working. The EU GDPR still uses the older prohibition model, so a UK employer hiring into the EU has two regimes; see our GDPR checklist for recruiters and the EU AI Act guide.

When is a hiring decision "solely automated" in the ICO's view?

This is the question Recruitment rewired spends most time on. The ICO spoke to more than 30 employers between March 2025 and January 2026. Most saw their tools as decision support because a hiring manager reviewed scores on a dashboard, but the ICO concluded that in practice managers rarely looked at lower-scoring candidates and that involvement was inconsistent even within one process. Its tests: human involvement must be active, "not just a token gesture or 'rubber stamp'"; the person must be able to influence the outcome before it takes effect and have the authority, discretion and competence to change it; and it "must be applied to every candidate, not just those who score highly." Designing or configuring the model does not count, because it happens too early to change a specific outcome.

The ICO's own use cases:

  • A fit score with an automatic rejection email below a pass mark: automated decision-making.
  • A ranked list where the manager invites the top candidates and rejects the rest without considering them: also automated decision-making for the people not interviewed.
  • Appointment or rejection after a face-to-face interview: not solely automated.
  • Mandatory screening questions a human set, such as right to work in the UK or a required qualification, with a "no" leading to rejection: not automated decision-making, because the tool only carries out an action a person predetermined.

If your process falls on the automated side, the ICO says you must either apply the Article 22C safeguards or change the process so there is meaningful human involvement in each decision about each candidate. Developers are told to review how their tools are designed and marketed.

What must employers tell candidates, and when?

The ICO's transparency findings separate three duties that employers tend to merge: the right to be informed when data is collected (Articles 13 and 14), the right of access (Article 15), and the Article 22C safeguard of information about each solely automated decision. Meeting one does not meet the others. Where automated decisions have significant effects, candidates need "meaningful information about the logic involved" and the likely consequences; for recruitment the ICO says to explain how the tool is used to decide, how accurate it is, and what safeguards exist, without needing to let anyone game the system.

What it found falling short: generic notices saying screening "may be automated," information that did not match practice, and employers sending candidates to the tool provider's privacy policy, which it calls poor practice. Only two employers offered human intervention, and then as a general dispute route rather than an Article 22 right. Good practice it saw: recruitment-specific privacy notices before application, information during and after the process, decision-specific explanations in rejection emails, and simple forms to ask for human review or contest a decision. For wording, adapt our AI disclosure templates and the UK notice in recording interviews under UK GDPR; candidate access requests are covered in candidate data access requests.

Which lawful basis does the ICO expect for AI in hiring?

Per the report's lawful basis section: legitimate interests is likely to be the most appropriate basis, with a legitimate interests assessment the ICO expects you to complete even though the law does not require one on paper; consent is unlikely to be appropriate at most stages because candidates may feel pressure to agree; contract fits only after an offer is accepted; and recognised legitimate interests "cannot" be used for automated decision-making. Special category data needs both an Article 22B condition and an Article 9 condition. The balancing test itself is set out in legitimate interest in recruitment.

What does the ICO expect on bias monitoring?

The audit report notes that many providers measured adverse impact with the "four fifths rule" as a minimum threshold (explained in the four-fifths rule) and warns that inferred demographics cannot do this job reliably. Its alternatives: collect demographic information directly, for example through an optional survey after the process, with a lawful basis for it. Recruiters should ask providers for test results and evidence of fixes before buying and regularly after. In 2026 the ICO added that it expects employers to keep asking developers about their bias testing, to run their own trials, and it cited monthly bias reviews as good practice. Discrimination itself is governed by the Equality Act 2010, which applies alongside the data protection rules.

What should recruitment agencies check?

  • Who is controller for what. The ICO says a provider is a controller if it uses the data it processes for you to develop a central model it deploys to all its customers. Agencies that build candidate databases from scraped profiles take on controller duties of their own, including notice within one month where data was not collected from the person.
  • Who tells the candidate. The audit found contracts in which neither party took responsibility, so candidates were told nothing. Name the party in the contract and check it happens.
  • Client decisions on your shortlist. If a client rejects candidates from an AI-ranked shortlist without reviewing them, the ICO's use cases suggest that is automated decision-making; agree who provides the safeguards.
  • Agency regulation. Conduct rules for agencies sit in separate law; see the Conduct of Employment Agencies Regulations. For the US position of agencies, see AI hiring laws for staffing agencies.

Checklist for UK recruiters and agencies

  1. List every tool that scores, ranks, filters or rejects candidates, and mark each decision point as solely automated or meaningfully human-reviewed, using the ICO's use cases.
  2. For each solely automated significant decision, put the Article 22C safeguards in place: an explanation, a route to make representations, a named human reviewer, and a contest process, with a response time.
  3. Check no solely automated significant decision uses special category data unless an Article 22B condition is met, and none relies on recognised legitimate interests.
  4. Where you rely on human review instead, apply it to every candidate at that stage and record it.
  5. Complete or refresh the DPIA before use, with your DPO involved from procurement.
  6. Document the lawful basis and a legitimate interests assessment for each processing activity.
  7. Publish your own recruitment privacy notice; do not point candidates to the vendor's.
  8. Get the provider's bias test results, run your own checks on real outcomes, and set a review interval.
  9. Fix controller and processor roles, processing instructions and notice responsibility in the contract.
  10. Set retention periods for AI outputs and records; see AI hiring recordkeeping requirements.
  11. Re-check the ICO's automated decision-making and recruitment guidance when the final versions are published.

How does this compare with the EU and the US?

In the EU, recruitment AI is high-risk under the AI Act, with deployer duties applying from 2 December 2027, on top of the GDPR's Article 22 (see the EU AI Act guide). In the US there is no single federal AI hiring law; cities and states such as New York City, California and Colorado regulate tools with their own definitions, compared in what an automated employment decision tool is and summarized in AI hiring laws by state. The UK approach is data protection first: the ICO regulates how personal information flows through the tool, and whether a decision is solely automated decides which safeguards apply.

Questions people ask

Does the ICO allow AI to reject job applicants automatically?

Yes, with conditions. Since 5 February 2026 the UK GDPR, as amended by the Data (Use and Access) Act 2025, permits significant decisions based solely on automated processing if the employer puts the Article 22C safeguards in place: information about the decision, a way to make representations, human intervention and a way to contest it. Stricter limits apply where special category data is used, and the ICO's March 2026 Recruitment rewired report says many employers already making solely automated decisions do not yet have those safeguards.

When does a hiring manager's review count as meaningful human involvement?

In the ICO's view, only when a person can actually influence the decision before it takes effect and has the authority, discretion and competence to change it, and when that review is applied to every candidate at that stage, not just the high scorers. A manager who only glances at low-scored candidates before rejecting them is rubber-stamping, so the decision is still solely automated.

Do I need a DPIA before using an AI recruitment tool in the UK?

Almost always. The ICO's 2024 audit report says DPIAs are likely to be required by law for AI recruitment tools, its key questions for recruiters put the DPIA first, and its 2026 report says that where automated decision-making is involved a DPIA is required. Do it before use, ideally at procurement, and involve your data protection officer early.

Is the ICO's guidance on AI in recruitment final?

Partly. The November 2024 audit outcomes report and the March 2026 Recruitment rewired report are published reports. The updated guidance on automated decision-making and profiling was still marked draft in November 2026 after a consultation that closed on 29 May 2026, and the recruitment and selection guidance was listed as in drafting for Winter 2026.

Can a recruitment agency rely on its AI vendor's privacy notice?

No. The ICO found that pointing candidates to the tool provider's privacy information is poor practice because it suggests accountability lies with the provider. Contracts should say which party informs candidates, and candidates should receive the employer's or agency's own privacy information.