AI tools promise faster pre-selection, better candidate outreach, and less administrative effort. At the same time, regulatory pressure in the EU is increasing: The EU AI Act (Artificial Intelligence Act) is coming into effect in stages and also affects applications in the HR sector. For Austrian employers, this means: By August 2, 2026 at the latest, recruiting processes must be set up in such a way that they are transparent, fair, and documentable – and for particularly sensitive systems (e.g., automated suitability assessments), further obligations apply from August 2, 2027 relevant.
In this article, you will receive a practical checklist: Which AI in recruiting is considered "delicate," which deadlines are important, which questions you should ask your provider – and which organizational measures are worth implementing in 2026, even if you do not (yet) have to meet all high-risk rules.
Why AI in recruiting is now a compliance issue
Recruiting is not just any process: Decisions regarding access to employment affect fundamental rights, equal opportunities, and often sensitive personal data. AI can be helpful in this regard – but it can also amplify risks, for example, if training data is biased or if a scoring model weights factors in an indirectly discriminatory way. This is precisely where the EU AI Act comes in: It classifies certain AI applications according to risk and links them to obligations for providers and users ("deployers").
For Austria there's an additional practical pressure: the labor market remains tight in many areas (skilled workers, healthcare, IT), and companies are trying to speed up processes. At the same time, candidates' expectations for transparency and fairness are rising. If you use recruiting AI, you should therefore be able to answer not only 'May I do this?' but also 'Can I explain it, substantiate it, and be accountable for it?'
AI Act in three dates: What applies when?
- Since 2 February 2025: Certain parts (including Chapters I/II) are already in effect — including obligations around AI Literacy (AI competence) and bans on certain practices.
- From 2 August 2026: The AI Act is generally applicable (standard effective date).
- From 2 August 2027: Certain rules for classification as “High-Risk” (Art. 6(1) and related obligations) will take effect later – relevant for many recruiting use cases that fall under “Employment, workers management and access to self-employment”.
Which AI in recruiting is particularly risky?
In everyday use the range runs from “harmless” to “highly sensitive”:
- Less critical: AI text suggestions for job ads, summaries of interview notes, internal HR FAQ chatbots (depending on the setup, still consider data protection/transparency).
- Medium risk: Matching tools that map profiles to requirements, automatically rank candidates, or generate “recommendations”. This can trend toward High-Risk – depending on whether it effectively prepares selection decisions.
- High risk: Systems that assess suitability, reliability, or performance (scoring), video/audio analysis, emotion recognition, automated rejection/shortlisting without sufficient controls. You should apply especially strict scrutiny here.
Practical rule of thumb: The closer the system is to a \"decision on acceptance or rejection the more you need proper documentation, human oversight, and a transparent process.
Checklist 2026: How to prepare yourself in a structured way
1) Inventory: Where do you actually use AI in HR?
- List all tools and features (ATS, applicant management, CV parsing, matching, chatbots, assessment platforms).
- Document for each tool: purpose, types of data, influence on decisions (informative vs. decisive), provider, hosting (EU/Non-EU), interfaces.
2) Risk scan: What could be discriminatory or non-transparent?
- Which attributes are used (direct/indirect)? Could they disadvantage protected groups?
- Are there 'proxy' effects (e.g., postal code as an indirect indicator)?
- How explainable is the result for HR and for applicants?
3) Human-in-the-loop: Clear rules for human oversight
- Define which decisions must never be made fully automatically (e.g., automatic rejection).
- Specify review steps: Who reviews, according to which criteria, and how is it documented?
- Create escalation paths (e.g., if the tool shows 'suspicious' patterns).
4) Data & privacy: ensure GDPR is properly considered
- Check the legal basis, information obligations, retention periods, and data processing agreements.
- Minimize data: only use what you really need for the purpose.
- For sensitive setups: consider a Data Protection Impact Assessment (DPIA).
5) Vendor questions: what you should get in writing
- Which data was used for training (to the extent possible)?
- What measures are in place to counter bias? Are there test reports, benchmarks, or audit information?
- How are logs/decision rationales stored? Can you trace decisions?
- How does human oversight work in the tool (override, justification fields, versioning)?
- Where are data processed (region), and who is the subprocessor?
6) AI literacy: Upskilling the HR team (relevant since 2025)
AI competence doesn't mean everyone has to be a data scientist. It's about practical skills: recognizing limits, understanding bias risks, not accepting results blindly, and being able to explain the key terms. Plan short trainings for HR and managers (e.g., 60–90 minutes), including case examples from your process.
7) Documentation: Capture the 'why'
- Why are you using the tool? What problem does it solve?
- Which risks have been identified — and how are they mitigated?
- Which KPIs do you monitor (e.g., hit rate, dropout rates, diversity indicators without personal identifiers)?
Concrete examples: What Austrian SMEs often overlook
- Job ad optimization: AI can unintentionally "gender" or de-gender phrasing, or exclude certain target groups. Use checklists and peer review.
- CV parsing: If parsing performs worse on atypical CVs, a structural disadvantage can arise. Test with realistic examples.
- Automated ranking: A ranking may appear objective but is often just a weighting of assumptions. Make the weightings transparent and review them regularly.
Roles under the AI Act: providers, operators ("deployers") and HR practitioners
In practice, many companies use an external applicant tracking system or an assessment platform. The allocation of roles is important:
- Provider (Provider) entwickelt oder stellt das AI-System bereit. Er muss – je nach Einstufung – technische Anforderungen erfüllen (z. B. Risikomanagement, Daten-Governance, Dokumentation).
- User/Operator (Deployer) setzt das System im Unternehmen ein. Für HR heißt das: You must ensure that the tool in its specific use is responsibly betrieben wird (menschliche Kontrolle, passende Prozesse, Information der betroffenen Personen, Monitoring).
Even if the provider advertises “AI Act-compliant”, employers are still left with the question: Does the tool fit your process?The same system can be merely supportive in one setting (lower risk) but in another effectively decide on rejections (higher risk). Therefore, it is worthwhile in 2026 to define a clear process — before simply enabling features.
Transparency in recruiting: What applicants should know
Regardless of the legal technicalities, applicants increasingly expect clarity about whether and how AI is used. That reduces friction and can improve the candidate experience. Practical measures:
- Short note in the privacy policy or on the application form:Which automated tools are used (e.g., CV parsing, matching)?
- Contact option for inquiries (HR address) and a clear statement that decisions are not made solely by automated processes.
- Internal guideline for recruiters: How are AI results used — and when are they ignored?
Austria special: equal treatment, works council, and practical pitfalls
In Austria, when using AI in recruiting you should also consider the "classic" issues that often come up too late in projects:
- Equal treatment: Check whether criteria or data points could indirectly lead to disadvantages. This applies particularly to rankings, automated preselection, and standardized assessments.
- Works council: If new systems are introduced that affect employees (e.g., internal mobility, performance or potential evaluations), co-determination may become relevant. Even if recruiting appears external: processes that later move into "workers management" (e.g., internal transfers) should be aligned early.
- Documentation obligations in everyday work: The most important lever is not a 40-page concept, but a continuous audit trail: Which decision was made, based on which information, and how was AI used?
Mini-template: AI tool fact sheet (copyable for HR)
- Name/Provider: …
- Application area: Stellenanzeige / CV parsing / matching / assessment / interview support
- Decision influence: nur Empfehlung / Ranking / automatische Aktion
- Data sources: CV, cover letters, profiles, test results, notes
- Human Oversight: Who reviews? Which overrides are mandatory?
- Bias checks: Which tests? How often? Who documents?
- GDPR: DPA in place? Subprocessors? Retention periods?
FAQ: Short practical answers
"May I use AI to sort applications?"
Yes — but the more the system influences selection decisions, the more important human oversight, transparency, and documented risk mitigation measures become.
“Is it enough if the provider says they are compliant?”
No. You also need process rules (human-in-the-loop), an auditable implementation and written proof of how you manage risks.
“What's the most sensible immediate step for 2026?”
A tool inventory plus a standardized fact sheet for each system. That lets you prioritize risks and explain decisions later.
Further reading on jobspot.at
- Austrian labor market in spring 2026: What job seekers can do strategically now
- Tips for the job interview
- Becoming a web developer: How to get into the IT industry
Conclusion: start in 2026, refine in 2027
Even though many high-risk obligations for certain AI systems come into effect later: those who start in 2026 with an inventory, human oversight, clean documentation and AI expertise reduce risks — and build trust with applicants. For Austria this is especially important: in a tight labor market, those who recruit fairly, quickly and transparently win.
CTA: Do you want to find suitable jobs or apprenticeships right now? Browse the latest listings on jobspot.at and set up search profiles – that way you won't miss a chance when the next suitable job goes online.
