Recruiting & HR: the EU delay is law and the CV stopped working

Recruiting and hiring now runs under two European rule dates and one fact about applications. On 24 July 2026 the EU's delay of the heavy duties on hiring software (bias testing, records, human oversight) became law, pushing them to December 2027, while the duty to tell a candidate they are dealing with an AI took effect on 2 August. In the US most employers screen by software, and recruiters believe most applications are written by it. The compliance file moved to 2027; the disclosure, and the decision, stay with the employer today.

Europe: the heavy rules moved, disclosure landed

When Era Haus last looked at recruiting and hiring, EU governments had just voted to postpone the AI Act's "high-risk" rules for hiring software. That vote is now law: Regulation (EU) 2026/1744 was published in the Official Journal on 24 July 2026 and took effect on 27 July. A tool that screens, ranks or scores candidates for a European employer has until 2 December 2027 before bias testing, documentation, human oversight and registration apply.

One duty kept its date; an older rule still applies. Since 2 August 2026 a candidate who chats with a bot or sits an AI-run first interview must be told they are dealing with a machine; law firms Gibson Dunn and Freshfields confirm the date held. And GDPR Article 22, Europe's older data-protection rule, still bars a rejection made purely by software: a person must hold real say, and signing off on a score does not count. A vendor's "AI Act ready" badge is a sales line for the next 16 months; the disclosure and the human decision are the employer's obligations now.

California moved the same way. At the end of August its legislature passed AB 1883, banning employers from using AI to judge emotion from a person's face, voice or tone, enforceable by the worker or a public prosecutor at up to $500 per violation (HR Dive, 31 August 2026); it awaits the governor's signature. The EU AI Act already prohibits emotion inference at work, so interview products that score "enthusiasm" from a video are losing legal ground in Europe, and in California for staff.

Screening is the default, and the CV stopped working

A survey of more than 1,000 US human-resources and recruiting leaders by Paylocity, a payroll and HR software vendor, found 91% use AI somewhere in recruiting and 67% use it to screen CVs (HR Dive, 27 August 2026). Fieldwork undated, and the vendor has a product to sell.

A second vendor survey, from MyPerfectResume, a US CV-building service, holds the uncomfortable number: 65% of business leaders say the software rejects candidates before any human looks, and nearly half say it has rejected someone who would have advanced under human review (HR Executive, 27 August 2026). The other side is automated too: 71% of the recruiters Paylocity surveyed believe more than half the applications they read were written with generative tools.

Argentina is behind on the same path. The job board Bumeran found 46% of HR specialists there use AI in selection, mostly to filter profiles by keyword (Infobae, 20 August 2026). In late August it ran an online fair with more than 9,000 vacancies where its own AI assistant built each candidate's profile from the uploaded CV (Fortuna, 24 August 2026). When profile and filter are both machine-made, the CV carries less information for both sides.

The money is following. On 1 September the trade outlet HR Tech Feed reported one purchase that shows it. Recruiter.com, a US recruiting-services company, bought Feenyx, an interview platform sold on skills tests and on checking that the candidate on screen is real and unassisted by AI. Interview software now replaces the CV and checks the person behind it.

Randstad automated the task and kept the people

Randstad, the Dutch group and world's largest staffing company, reported second-quarter results on 22 July 2026 and described its AI gains as a lower cost to serve: roughly 50% lower recruiting cost and 50% shorter time to fill on its digital marketplace, and, its finance chief said, "our intent is to keep the team stable". Those figures come from one transcript; treat them as the company's claim. That reverses the cut-then-rehire cycle the last piece documented, and it is the version a small employer can copy. Automate the task, keep the person.

What it means for you

If you employ in Europe, compliance is a 2027 problem and disclosure a today problem. In California, any tool that scores an employee's tone or facial expression is about to become a liability; the bill does not name applicants. Anywhere, the CV has stopped telling candidates apart. The advantage goes to the employer who asks for something a model cannot produce on the candidate's behalf, and who has a person read the reject pile.

What to do about it

First, add the disclosure now. If a chatbot answers candidates' questions or an AI runs a first-round interview for a European role, put one plain sentence in the chat window and the invitation saying so. Keep the record the last piece asked for (who decided, what the tool recommended, why you agreed or overrode it); it rests on GDPR today and the AI Act's oversight duty from 2027. Switch off any feature that scores emotion or tone, wherever you are.

Second, replace the CV screen with a short task. A 20-to-30-minute work sample, AI tools allowed and the candidate asked to explain their choices, tells you more than a CV a model wrote and a model read. It also fills a gap: a survey of 207 HR leaders in the US and UK by Talogy, an assessment company that sells hiring tests, found 78% struggle to assess AI skills (HR Dive, 26 August 2026).

Watch, do not act yet: fraud-detection interview tools. Every figure on how many candidates are fake came from vendor blogs and analyst forecasts. Buy verification once you have seen a fake yourself.