AI training: data quality is key
It is not just the amount of data that counts. Successful AI training requires, above all, high data quality. Insufficient data quality threatens to become an AI showstopper. This has regulatory reasons, such as Art. 10 AI Act. But above all, it has technical reasons, because the AI reproduces an extract of the training data. “Garbage in, garbage out”.
Data quality can have many facets:
- Anonymity of data
- Freedom from third-party intellectual property rights
- Utilization
- Correctness
- Representative for the specific application
- Completeness
- And many more
Ensuring the right data quality is a technical and legal task, similar to creating a technical service description. Aitava supports you in an interdisciplinary way.
Data exchange – often a balancing act
Striking the right balance when sharing data is a complex matter. On the part of the sharing company, the tension between confidentiality and the intended data transfer must be resolved. Unchecked data sharing can undermine the protection of know-how and trade secrets.
For the data acquirer, too extensive insight into data and its links to compliance risks can lead to increased product monitoring obligations. Competition authorities are closely monitoring how much access to data favors market power and abusive pricing. And data protection authorities are also taking a close look, especially if the type and scope of the acquired data enables profiling of individuals.
Best Practice: Data License Agreement
When planning a data exchange, the question arises as to how the own value-added share can be sustainably secured. Neither German nor European law recognizes data ownership. According to the current legal situation, economic data allocation can only be carried out on a contractual basis. So what the law has not regulated by legislation, the economy regulates among themselves at the contractual level. In this context, data license agreements are becoming increasingly important as they enable the parties involved to find flexible solutions for utilizing the data that is of interest to them. Figuratively speaking, data license agreements are legal drilling rigs that can be used to extract and use the ‘oil of the 21st century’ in a legally secure manner.
In addition, certain non-personal data may in future only be used and transferred on the basis of a contract with the respective user in accordance with Art. 4 para. 13, 14 Data Act. There are three basic approaches for the efficient introduction of data license agreements with users: the supply chain solution, the account solution and the pragmatic introduction of data licenses in side letters.
We draft and negotiate data license agreements on an ongoing basis. From one-pagers to 100-page documents.
AI as subject matter of a contract
Artificial Intelligence is becoming increasingly important in the economy and is more and more frequently included as an object of contract in business agreements. This not only concerns the purchase or licensing of AI software, but also the development, implementation and use of AI systems and their data and algorithms. A central issue in the inclusion of AI in contracts is the question of liability: Who is liable if an AI system malfunctions or makes unexpected decisions? How can I integrate AI base models into my product or SaaS services without adopting and being liable for any errors in the base model? A particular challenge lies in integrating the dynamic nature of AI into static contracts while maintaining the flexibility to respond to technological developments (e.g. model drifts). AI as a contractual object therefore requires a high degree of legal precision and technological understanding – and this is where Aitava likes to come into play!
It is to be expected that many participants in the AI value chain will in future be jointly and severally liable for AI damage – possibly even IT suppliers and end users. We show you how to optimally allocate such risks today through appropriate liability or recourse agreements. And cover them with suitable AI insurance. Aitava gives you a clear overview of current and future liability issues. We help you identify impending liability risks at an early stage and introduce customized protective mechanisms to shield you from potential risks.
New rules of the game: the Data Act and the AI Act
The Data Act will apply from September 12, 2025. In essence, it is a reorganization of the framework conditions for data access and data use. Real access rights to data are to be created for both the private and public sectors in order to boost data exchange. At the same time, significant obligations are to be imposed on data owners, product manufacturers and cloud providers. In the case of data license agreements, previously common clauses are to be partially prohibited.
The AI Act came into force on July 1, 2024 and is now to be implemented in stages by 2026. It has a significant impact on the design of AI contracts: On the one hand, companies are well advised to always keep Art. 25 AI Act (responsibilities along the AI value chain) in mind when drafting contracts. On the other hand, the specific obligations of the AI Art (in particular the AI prohibitions, transparency obligations and high-risk obligations) should always be considered when drafting contracts.
The Data Act and the AI Act fundamentally change the rules of the game for AI and data contracts. We help you to prepare for them at an early stage.
