EU AI Act's Design Impact Emerges
The EU's AI Act is moving into its compliance phase, with specific rules for designers of public services. "High-risk" systems like grant screening will require human oversight and user recourse. Even common tools like ChatGPT and Copilot face new regulations on output reliability and data protection when used in government workflows.
For public bodies and private entities providing public services, a key design-related mandate is the Fundamental Rights Impact Assessment (FRIA) required before deploying a high-risk AI system. This assessment goes beyond technical compliance to systematically evaluate and mitigate risks to privacy, human dignity, and non-discrimination, requiring a detailed description of the AI's processes, the people it will affect, and planned human oversight measures. The regulation explicitly links to existing accessibility laws. High-risk AI systems must comply with accessibility requirements outlined in the EU Accessibility Act, embedding an "accessibility by design" approach. This ensures that as government services automate, they do not create new barriers for people with disabilities, a concern actively voiced by groups like the European Disability Forum. For the science and technology sector, the European Research Council (ERC) has clarified that while researchers may use AI to brainstorm or edit grant proposals, the human author retains full responsibility for the content's integrity and originality. Funding bodies themselves are approaching AI evaluation cautiously; the ERC currently does not use AI for initial proposal screening or review, citing concerns about potential bias. The Act's rollout is staggered, with rules for high-risk systems becoming fully applicable on August 2, 2026. To facilitate innovation within these new constraints, EU member states are required to establish "AI regulatory sandboxes" by that date, offering controlled environments for companies to test new technologies with regulatory guidance. Non-compliance carries severe financial penalties, designed to be dissuasive. Fines for using prohibited AI practices can reach up to €35 million or 7% of a company's total worldwide annual turnover, whichever is higher. Supplying incorrect or misleading information to regulators can result in fines of up to €7.5 million or 1% of turnover. Across Europe, early public sector AI adoption offers practical examples. In Kortrijk, Belgium, a multilingual virtual assistant helps citizens access services, while a project in Nicosia, Cyprus, co-designed a mobility assistance tool with disability organizations to build trust and ensure utility. These cases highlight a focus on tangible public benefits and stakeholder engagement ahead of the Act's full implementation.
