AI in Recruiting: Where It Helps, and Why It Becomes High-Risk in December 2027

HR is full of text and routine: job ads, rejections, onboarding plans, answers to the same questions again and again. At the same time, HR makes decisions about people. That tension defines where AI helps and where it becomes dangerous.

The red line

The AI Act classifies systems used for employment and recruitment as high-risk AI (Annex III). That covers tools that filter or score applications or rank candidates. Under the Omnibus Regulation (EU) 2026/1744, the duties for these systems apply from 2 December 2027R. Providers and deployers will then need risk management, human oversight and documentation, among other things. If you introduce a scoring tool today, you are building on an obligation that will certainly come.

Independently of the AI Act, Art. 22 GDPR applies: decisions with legal effect may not be based solely on automated processing. And Germany's General Equal Treatment Act prohibits discrimination, including when it comes from a model.

Where AI helps safely today

• Job ads: In an interview quoted by Quartz, study author Anders Humlum names writing job ads as an example of tasks chatbots noticeably speed up.
• Feedback processes: According to Zapier, Miro raised participation in peer feedback from 50 to 93 %H by automating reminders and forms.
• Onboarding: Checklists, welcome emails and calendar invites are generated from a template.
• Internal FAQ: An assistant answers questions on leave, expenses or benefits from approved documents.

We have not yet seen an independent German case study with reliable figures on AI in recruiting. If you know one, write to us.

The workflow: a job ad with a discrimination check

The works council

In 2024 the Hamburg Labour Court ruled that merely allowing employees to use private ChatGPT accounts does not require works council approval. If the employer provides a tool itself, most law firms consider co-determination under Section 87(1) No. 6 of the Works Constitution Act to apply. Involve the works council early.

How to start

1. Write it down: AI drafts, people make personnel decisions.
2. Start with job ads and onboarding material.
3. Build the discrimination check into a fixed prompt step and have HR review it.
4. Do not introduce candidate scoring until high-risk duties, a data protection impact assessment and a works agreement are settled.

Sources

1. RCloud Security Alliance: EU AI Act's High-Risk Deadline: Deferred, Not Cancelled
2. RGibson Dunn: EU AI Act Omnibus Agreement, Postponed High-Risk Deadlines
3. RRegulation (EU) 2016/679 (GDPR), Art. 22
4. RGleiss Lutz: Hamburg Labour Court on ChatGPT and works council rights
5. MQuartz: AI is storming workplaces and barely making a difference, study says
6. HZapier Customer Stories (Miro)