EU Rules Tighten the Reins on AI in the Workplace as Courts and Unions Push Back
Published on 08/25/2026 at 21:41 | Redaktion boerse-global.de
The clock is ticking for employers who have embraced artificial intelligence in hiring, pay reviews, and dismissals. While the technology promises speed and efficiency, a wave of regulatory deadlines and courtroom battles is forcing a rethink of how automation fits into people management.
From next year, a fresh set of obligations lands on companies deploying AI in human resources. The EU's AI Act, which classifies such systems as high-risk, has already banned certain practices—including emotion recognition at work—since 2 February 2025. Operators and developers face additional duties from 2 August 2026, with full compliance for high-risk systems required by 2 December 2027. Alongside this, the General Data Protection Regulation (GDPR) sets a hard boundary: under Article 22, fully automated decisions leading to dismissal are off the table, and a human must always have the final say.
That legal scaffolding is being tested in real cases. In the United States, job applicant Erin Kistler has taken Eightfold AI to court, alleging that its screening software—which draws on data from more than one billion workers and rates candidates on a scale from 0 to 5—lacks transparency and offers no meaningful way to challenge or correct automated scores. Tech giants are also in the crosshairs: Meta faces criticism over AI-driven terminations linked to parental leave or illness, while IBM has been accused of age discrimination in its use of such tools.
German courts are adding their own clarifications. On 15 January 2026, the Regional Labour Court of Schleswig-Holstein ruled (case number 4 Sa 62/25) that a read receipt does not automatically prove that a job application, including attachments, was legally received. Under the General Equal Treatment Act (AGG), a candidate's formal status only arises when the application arrives through the designated channel. The Federal Labour Court is set to hear an appeal on 18 March 2027.
Despite the legal friction, appetite for automation remains strong. The Institute for Employment Research (IAB) calculates that 86 percent of tasks performed by HR administrators could be automated. A Kienbaum survey finds 35 percent of companies already use AI in recruitment, with another 26 percent planning to do so. Among 1,500 workers polled by 3Gem and Adobe, 67 percent say AI saves them up to eight hours a week.
Yet the push for speed is colliding with worker representation. The federal government, following a coalition committee decision on 2 July 2026, wants to curb works council co-determination rights over IT systems under § 87 (1) No. 6 of the Works Constitution Act (BetrVG) to accelerate AI rollouts. At the same time, Berlin aims to lift the continuing-education participation rate to 65 percent by 2030, anticipating not mass job losses but a significant shift in task profiles.
Pay-setting is another frontier. Simulations by researchers at the University of Bielefeld suggest that learning algorithms on online platforms can produce collusion-like low wages without any explicit coordination between employers—a finding with implications for competition policy. In the US, 78 percent of companies already use AI assistance when deciding on salary increases.
The legal profession remains split. Some firms refuse AI in client work over liability concerns, while others integrate it early into training. Startups like LegesAI are building infrastructure around local language models and a human-in-the-loop approach, aiming to anonymise sensitive client data in compliance with privacy rules.
