Compensation is perhaps the most sensitive area of AI in HR. Salary, bonus and equity decisions directly affect workers' livelihoods, and bias in compensation AI compounds over years — a lower starting salary carries through every subsequent raise. For the EU AI Act this makes compensation AI evidently Annex III point 4(b) high-risk. And as the European pay transparency directive becomes hard from 2026, new obligations stack on top.
This post explains where AI sits in compensation, why the classification question is sharper here than in other 4(b) deployments, and what HR teams must arrange in 2026.
Where AI is deployed in compensation
The modern compensation stack increasingly covers:
- Salary benchmarking AI — Mercer Mettl, Payscale, Figures.hr, Ravio: AI aggregates of market data per role, geography, experience
- Raise and bonus recommendations — SAP SuccessFactors Compensation, Workday Compensation, HiBob: AI suggests raise percentages based on performance, market data and budget
- Equity and stock decisions — especially at scale-ups: AI suggests grants based on role, tenure and performance
- Pay equity audits — Trusaic, Syndio: AI analysis of pay gaps over demographic dimensions
- Total Rewards optimization — AI suggests benefit packages per person based on demographics and behavior
- Skills-based pay — AI-driven pay bands tied to inferred skills via Talent Intelligence Hub-like systems
Not every compensation tool is high-risk by definition, but the combination of compensation + AI has a specific risk profile HR teams underestimate.
Why compensation AI more sharply hits 4(b)
Compared to other 4(b) deployments, compensation AI has three properties that make the classification decision sharper:
- Material impact — pay decisions are more direct and quantifiable than other worker management decisions. A lower salary is a measurable effect.
- Cumulative impact — pay bias compounds. A 5% lower starting salary means hundreds of thousands of euros difference over a 30-year career.
- Protected categories interaction — pay equity directly affects gender, ethnicity, age. Anti-discrimination law stacks on AI Act.
In 2026 the EU Pay Transparency Directive comes on top. Employers with 250+ employees must report pay gaps annually from 2026, and candidates/workers can request pay information. If AI helps feed compensation decisions, you are required to explain how.
When is compensation AI within 4(b)
- Pure market benchmarking for reference — if output is only consulted and feeds no specific worker decisions: lighter. Often outside 4(b).
- AI suggests raise percentages for individual workers — directly within 4(b).
- AI-driven calibration between managers — affects pay outcome per worker. Within 4(b).
- Skills-based pay with AI-inferred skills — if inferred skills determine compensation level: within 4(b).
- Pay equity audits without individual recommendations — usually outside 4(b), but within GDPR.
- AI-suggested promotions or bonus outcomes — directly within 4(b).
Step-by-step for compensation AI dossier
Treat compensation AI as heaviest 4(b) category
No low-effort dossier. The combination of AI Act, GDPR, Pay Transparency Directive and anti-discrimination law makes compensation AI your most scrutiny-sensitive deployment.
Split aggregate from individual
Pay benchmark tools that only show market aggregates sit differently from tools suggesting per-worker raise percentages. Document per use case.
Build dossier via HR AI Evidence Pack under 4(b)
The HR AI Evidence Pack has a section for compensation context. Per AI tool fill in impact analysis and oversight procedure.
Test your AI against the Article 6(3) filter
Interactive self-assessment, updated for the Commission guidelines of 19 May 2026. 9 steps, personal report with reasoning, vendor questions and next steps.
Frequently asked questions about compensation AI and the AI Act
Practical questions for HR on Annex III point 4(b) and pay decisions.
What to do now
For HR, finance and compliance teams with compensation AI: treat this as priority-1 within your 4(b) trajectory. Tool inventory this month, FRIA with cumulative impact analysis within 60 days, Pay Transparency Directive integration in your 2026-2027 roadmap. Document via the HR AI hub and the HR AI Evidence Pack.