Supporting Assurance Case Development Using Generative AI

As the use of cyber-physical systems in safety-critical domains continues to rise, assurance cases have become a widely adopted approach for justifying the safety of these systems. During assurance case development, errors can occur such as logical fallacies and argument incompleteness which can lead to the deployment of unsafe systems. Methods such as Eliminative Argumentation have been proposed to improve confidence in assurance cases by identifying potential doubts in the argument (called defeaters) and arguing that they have been appropriately mitigated; however, using these methods does not guarantee that engineers will identify all relevant defeaters. In this paper, we propose our vision for using generative artificial intelligence to aid in the identification of defeaters in assurance cases to improve their reliability.