European AI Act: Penalties take effect on 2 August 2025

The European Union’s Artificial Intelligence Act (AI Act), the world’s first comprehensive AI regulation, reaches a major milestone on 2 August 2025. This date marks the start of several key obligations for companies, governments, and AI providers across the EU, including penalties for non-compliance. SD Worx, the leading European HR solutions provider, summarises what this means for organisations and their workforce. 

The AI Act came into effect on 2 February earlier this year, introducing a uniform legal framework for artificial intelligence within the EU. Although many provisions will not take effect until 2026, a new phase will start on 2 August 2025 with a focus on three areas:

1. Penalties for non-compliance
2. Obligations for general purpose AI (GPAI) models
3. Governance and oversight at national and European level

From 2 August 2025, fines may be imposed for breaches of obligations already in place. Since 2 February this year, AI systems with unacceptable risk have been banned and companies must ensure their employees are AI literate. Those who fail to respect this risk fines of up to EUR 35 million or 7% of their global annual revenues. The European Union expects member states to set their own effective, proportionate and dissuasive penalties. In doing so, they must take into account the situation of SMEs and start-ups so that the fines do not jeopardise their economic viability.

General Purpose AI (GPAI) models that will be marketed in the European Union from 2 August 2025 are subject to specific obligations. The European AI Office published the final version of codes of practice on 10 July 2025. Providers of such GPAI models must, among other things, prepare technical documentation, respect copyrights and provide transparency on training data used. 

GPAI models are designed to perform a wide range of tasks. They are trained on huge amounts of data and are versatile. The most common example are large language models (LLM), such as the generative language model GPT-4o integrated into ChatGPT. Providers of GPAI models already on the market in the European Union before 2 August 2025 will still have until 2 August 2027.

The AI Act introduces a framework with implementation and enforcement powers at two levels. 

Nationally, each EU member state must appoint at least one market surveillance authority and one notifying authority by 2 August 2025. These authorities are responsible for respectively monitoring AI systems and notifying independent conformity assessment bodies. Member states must publish information on national authorities and how to contact them by 2 August 2025.

At the European level, supervision will be coordinated by the AI Office and the European AI Board. An Advisory Forum and a Scientific Panel of independent experts will be set up.

The AI Act has direct implications for how AI is used in recruitment, performance management, workforce analytics and employee monitoring. HR leaders must ensure that any AI tools used in these areas are transparent, fair, and compliant with the new rules.

• Fairness and non-discrimination: AI systems used in hiring or promotion decisions must be explainable and free from bias. HR teams should review their tools and vendors to ensure compliance.
• Employee trust and transparency: employees will have greater visibility into how AI systems affect them, whether it’s in scheduling, evaluations, or workplace surveillance. This is an opportunity for HR to build trust by clearly communicating how AI is used and how employee data is protected.
• Vendor accountability: HR departments using third-party AI tools must ensure those providers meet the new transparency and documentation requirements. Contracts and procurement processes may need to be updated accordingly.
• Training and change management: as AI governance becomes more formalised, HR will play a key role in training managers and employees on responsible AI use and ensuring ethical practices are embedded in workplace culture.