Algorithms that are essential for the functioning of modern AI systems are neither protected under patent and copyright doctrine, nor under the regime of sui generis database rights. Inevitably, therefore, it is left to the parties of so-called AI-as-a-service contracts to privately assign rights of access and use. This system of private ordering is problematic in many respects: Not only does it leave algorithms widely unprotected against the risk of misappropriation. In addition, the imbalance of power between the parties in the AI-as-a-service industry is a fundamental threat to competition and innovation. More concretely, the currently dominant usance of assigning the providers exclusive rights of access and use of the algorithmic models developed by their customers’ data and AI training effort prevents the emergence and the efficient functioning of a “marketplace for algorithms”. As can be shown, this state of lock-in and disruption must be broken up and mended by defining and allocating a property right in the algorithm.
