Table of contents
The European Parliament and the Council of the EU reached a provisional political agreement on May 7, 2026, on the Digital Omnibus AI, a package of targeted amendments to Regulation (EU) 2024/1689. At the heart of the compromise is the most structuring deadline of the text for human resources departments and data protection officers: the Annex III obligations, which cover HR uses (recruitment, selection, promotion, contract termination, task allocation, and performance monitoring), are postponed by 16 months. The deadline shifts from August 2, 2026, to December 2, 2027. The postponement was approved in a plenary session in Strasbourg with a vote of 569 in favor, 45 against, on March 26, 2026, with the trilogue opening the same day. This delay occurs as 99% of CEOs anticipate that AI will lead to layoffs over the next two years, according to a survey published by Mercer at the end of May 2026 among nearly 1,000 executives.
99% of CEOs anticipate AI-related job cuts within two years
Among the approximately 1,000 executives surveyed by Mercer in 2026, only 1% do not foresee short-term AI layoffs - while the reported well-being of employees has dropped from 66% to 44% in two years.
A postponement formalized as job pressure is documented
The postponement occurs as labor market pressure on the AI subject accumulates in the most recent surveys. Among the panel of nearly 1,000 executives surveyed by Mercer at the end of May 2026, the share of employees reporting being fulfilled at work falls from 66% in 2024 to 44% in 2026. The Federal Reserve Bank of New York observed in the first quarter of 2026 an underemployment rate of 41.5% among young graduates for an unemployment rate of 5.7%, indicating that the difficulty no longer lies solely in accessing the market but in the quality of offered positions. On the cultural front, the symbolic episode of the month came from former Google CEO Eric Schmidt, who was booed during the graduation ceremony at the University of Arizona in May 2026 on the theme of graduates' role in building AI. Schmidt conceded that students' fears about job disappearance and a compromised future were "founded" (freely translated), with similar incidents reported at the University of Central Florida and Middle Tennessee State University.
Exact scope of the created void: Annex III point 4 and AI literacy
Point 4 of Annex III of the AI Act covers, in the current state of Regulation (EU) 2024/1689, two distinct categories of uses. Point 4(a) targets AI systems used for recruitment or candidate selection, including for publishing targeted job ads, analyzing and filtering applications, or evaluating candidates. Point 4(b) covers systems intended to make or inform decisions affecting the employment relationship: promotion, contract termination, task allocation based on behavior or personal traits, monitoring, and performance evaluation. All these uses are classified as high-risk by the regulation, and the application of substantive obligations - risk management, compliance assessment, technical documentation, transparency, human oversight, post-market surveillance - is postponed by 16 months. According to the analysis of the package published by the firm noze, the political agreement of May 7, 2026, was obtained under the combined pressure of the industry, national authorities, and consumer organizations, who argued the impossibility of completing compliance paths within the initial deadlines; only 8 Member States out of 27 had designated their competent national authorities by March 2026. The European Trade Union Confederation (ETUC) and industriAll Europe, for their part, denounced the concomitant weakening of the AI literacy obligation for providers and employers carried by Article 4 of the regulation: the standard shifts from the obligation to "ensure a sufficient level of AI proficiency" to the more vague "support the improvement of AI proficiency". "The few workplace references in the AI Act now risk being weakened by the Digital Omnibus" (freely translated), judges the ETUC.
Postponement ≠ regulatory void
The Digital Omnibus postpones by 16 months (from August 2, 2026, to December 2, 2027) the application date of high-risk obligations of the AI Act - it does not eliminate them. Risk management, compliance assessment, technical documentation, human oversight: these requirements remain intact. The pressure initially came from an infrastructure deficit: only 8 of the 27 Member States had designated their competent national authority by March 2026.
What remains: GDPR Article 22 and its jurisprudential limit
The postponement does not erase all protection. Article 22 of the GDPR remains fully applicable to HR processing based on an exclusively automated decision producing legal or significant effects on the person, at the forefront of which are recruitment, dismissal, or individual evaluation decisions. The Court of Justice of the European Union set this requirement for effective human intervention in its SCHUFA Holding (Scoring) judgment of December 7, 2023 (C-634/21): the mere nominal intervention of a human operator does not remove processing from the scope of Article 22; it must still have real discretionary power. The Court complemented this framework in 2025 by requiring useful information on the logic of automated decisions. The practical scope for legal departments and data protection officers is circumscribed: HR AI systems that do not operate a strictly automated decision, typically a scoring tool that feeds a collegial committee with effective discretionary power, remain outside the direct scope of Article 22, even though they would have been subject to Annex III point 4 obligations from August 2, 2026. This is precisely the scope that the SCHUFA judgment (C-634/21) left open for the AI Act: the additional 16 months formalized by the Digital Omnibus further delay the regulatory response that this text was to provide to HR managers and DPOs who are already deploying HR scoring tools.