AI hiring in 2027: What happens when a machine rejects your CV under the EU AI Act?

Published 15 November 2026. This is forward-looking reporting; official schedules and implementation details may change.

A candidate applies for a job and receives a rejection seconds later. Did a recruiter read the application, or did a system rank the applicant below an invisible threshold? Automated screening is not new. What changes is the coming European regulatory treatment of high-risk AI used in recruitment and other decisions that can substantially affect a person’s opportunities.

Why recruitment AI can be classified as high-risk

The EU AI Act distinguishes uses of artificial intelligence by risk. Certain systems used in employment—such as AI intended for recruitment and selection, including targeting advertisements, analysing and filtering applications, or evaluating candidates—fall within the high-risk category under Annex III. There are exclusions and qualifications, so not every automated spreadsheet or keyword search is automatically a prohibited high-risk system.

Under the revised EU timetable, obligations for stand-alone high-risk AI systems are due to apply from 2 December 2027. Certain product-embedded systems follow a later schedule. A website claiming all AI hiring becomes illegal that day is simply wrong: the law regulates specified high-risk uses; it does not enact a blanket ban on automated assistance.

What employers will need to do

The legislation places obligations on providers and deployers concerning risk management, data quality, documentation, logging, appropriate human oversight and other safeguards. Employment-related systems may create serious discrimination risks when training data reflect earlier hiring practices or proxies such as address, educational background or career gaps.

The phrase ‘human oversight’ must mean something operational. Can a person examine the basis of a recommendation? Is there an audit trail? Can a recruiter recognise when a model is confidently wrong? Is someone trained and authorised to override a recommendation? A nominal human approving every machine-generated ranking without review does not necessarily make the process fair.

What can a rejected candidate actually know?

Applicants should distinguish a moral intuition from an established legal remedy. The AI Act interacts with the GDPR and national employment and discrimination laws, each of which has particular scopes and conditions. The GDPR already includes protections concerning certain solely automated decisions with legal or similarly significant effects. The AI Act adds requirements for high-risk systems and transparency under specified circumstances.

The practical questions are whether AI was used, what role it played, what data were considered, how to correct errors and whether a meaningful review is available. An opaque score alone should not become a substitute for assessing the person’s actual capacity to do the job.

The historical analogy: from punch cards to invisible rankings

Herman Hollerith’s punched-card machines were celebrated for accelerating tabulation in the late nineteenth century. Computers made enormous populations legible to administrators by converting messy individual histories into standardised fields. Modern hiring models extend that tendency by turning candidates into numerical patterns.

The political question is not simply whether machines are biased. Human recruiters have their own prejudices and inconsistencies. It is whether decisions become more defensible because a machine processes them—or less defensible because nobody can say exactly why a person was excluded. In 2027 the answer may increasingly become a legal as well as ethical matter.

Sources and further reading

European Commission: AI Act and high-risk systems

EU AI Act implementation timeline

Council of the EU: June 2026 revised deadlines

Featured image: authentic documentary image and attribution record. A replica of Herman Hollerith’s tabulating machine at the Computer History Museum: a historical antecedent of automated sorting, not a modern hiring AI. The image was placed in a 1200 × 675 frame without cutting off the source.

Support DEPTH

Independent explanations grounded in primary sources take time. If this article helped you understand what is changing, you can support the work.