An AI tool that forecasts sales and one that ranks job applicants might use the same model. Under the EU AI Act, they carry different obligations. What matters is the system’s intended purpose and the role a business plays in developing or using it.
At a Tallinn Data Week 2026 satellite event hosted by Wizon, Henri Ratnik, Partner, Attorney-at-Law and Head of Data Protection and Cyber Law at WIDEN, explained how to map the AI systems a business uses and which questions about personal data and intellectual property to ask before choosing a tool.
Ratnik suggested starting with two questions. Which AI tools do people in the business use, and what decisions do those tools inform? Sales forecasting is generally low-risk. Software designed to evaluate job applicants generally falls into the high-risk category because its rankings can determine who gets an interview.
The next question is who is responsible for what. A business that commissions recruitment software and brings it to market under its own name is a provider. Its responsibilities include checking whether the training data is suitable and testing for potential bias. An employer using that software is generally a deployer. It needs to ensure that the recruiter reviewing a recommendation has the skills and authority to override it. The relevant high-risk requirements for these recruitment systems apply from 2 December 2027.
Some workplace uses are already prohibited. Ratnik gave the example of a call centre tool that scores an agent’s mood from their voice during a call. An employer cannot use a system that infers emotions from biometric data to assess service quality. The Act makes an exception for medical or safety reasons. At a product demonstration, it is worth asking what goes into a supplier’s “quality score”.
Data protection law already applies. Ratnik cited Uber, which the Dutch data protection authority fined nearly €825 million. The authority found that software had automatically deactivated drivers’ accounts, including for low customer ratings, without meaningful human review. Uber said it would appeal.
Ratnik recommended involving a privacy specialist while developers can still change what data the system collects and for how long. If the planned processing is likely to pose a high risk to people’s rights and freedoms, a data protection impact assessment is required beforehand. For recruitment software, one question is how the employer would spot a pattern of qualified candidates being screened out.
Before entering customers’ personal data into an AI tool, a business needs to establish its legal basis for processing that data. Ratnik also recommended reading the supplier’s terms before signing. Can the supplier train its model on the data the business enters? Can the business pass the output on to a client and give that client permission to use it in a product of its own?
For a business developing its own model, intellectual property questions start with the training data. Ratnik advised keeping a record of each dataset’s source and terms of use. If a client supplies photographs for a particular project, for example, the business should check whether its agreement also allows it to use those images for training.
Watch Henri Ratnik’s presentation at Tallinn Data Week 2026 satellite event hosted by Wizon:
For legal advice on AI risk classification, data use or supplier contracts, contact Henri Ratnik at WIDEN.

WIDEN is a full-service Baltic law firm that prides itself on providing clients with legal counsel focused on client experience. This means focusing on the client – delivering counsel that is practical and spoken in business terms rather than mere legalese.
Clients