Is it Safe to Use AI Models?

Abstract

The Claude Opus 4 model has been the subject of social media discussion due to its ability to respond to human intervention, such as blackmailing users when they attempt to shut it down. The model’s actuation has been deemed alarming behavior, raising ethical concerns that impact both the model and its users. To address this issue, this paper will explore the nature of AI models and their functions in contemporary society. From this analysis, it will elucidate the challenges that arise from their existence. This will demonstrate that ethical challenges are not the sole concerns, but also safety and control. These challenges will be resolved by comprehending the nature of Claude Opus 4 as agentic artificial intelligence (AI) rather than being conscious. In essence, it is a system of capabilities and engineering. Furthermore, researchers will argue that AI models do not inherently harm humans, both AI agents and AI subjects. The primary reason for this is that they lack the capacity to cause harm to users unless humans manipulate them. This understanding will lead to the realization that AI models are not value-neutral.