The CARAT project investigates how certification can be used as a key regulatory instrument for managing risks and ensuring the trustworthiness of artificial intelligence (AI) systems. The research project analyses the extent to which effective certification processes can make a decisive contribution to minimising risks such as discrimination, lack of transparency, data protection violations, security problems, and ethical challenges. The overarching goal is to develop a comprehensive legal and technical framework that promotes compliance, innovation, and international harmonisation of AI regulation standards.

CARAT is an interdisciplinary project within the framework of a German-Brazilian cooperation between Saarland University and the University of São Paulo. On the German side, it is funded by the German Research Foundation (DFG), while the Brazilian partners are funded by the Brazilian Federal Agency for Support and Evaluation of Graduate Education (CAPES). At Saarland University, the project is led by the Chairs of Prof. Dr. Georg Borges and Prof. Dr. Holger Hermanns. On behalf of the University of São Paulo, Prof. Dr. Juliano Souza de Albuquerque Maranhão is leading the project.

The CARAT team conducts detailed analyses of existing and new certification methods under key regulatory frameworks, particularly the European AI Act and the General Data Protection Regulation (GDPR). By examining explicit and implicit references to AI certification in these legal regulations, the project identifies how certification can strengthen compliance with the law and increase public trust. In addition, CARAT explores innovative certification models that are suitable for both voluntary and mandatory regulatory contexts, while enabling flexibility and consistent adherence to ethical and legal standards.

From a technical point of view, the CARAT project focuses on identifying and evaluating state-of-the-art methods, best practices, and international standards required for robust AI testing and verification. Technical challenges, such as the so-called ‘black box’ effect, where the internal decision-making processes of AI systems are often not transparent, are at the forefront and underscore the need for comprehensive external testing and auditing procedures.

CARAT focuses in particular on the following key questions:

• What are the legal implications and practical effects of certifying AI systems?
• How can regulatory frameworks effectively promote the responsible and transparent development and use of AI?
• What technical standards and industry-specific criteria should certification systems include to ensure global applicability and acceptance?
• Can common certification principles be established between Europe and Brazil, and what are the key differences?

The expected results of the CARAT project include comprehensive legal assessments, concrete proposals for voluntary and mandatory certification frameworks, and clear technical guidelines for compliance with legal requirements. The project results will be widely presented at international symposia and summarised in a final comprehensive publication aimed at policy makers, industry representatives, academic researchers, and the general public.