In a recent paper including authors from WSTnet lab Oxford Internet Instute, the emerging perceptions of relationships with AI systems are explored.

As AI systems become more personalized and socially interactive, people are increasingly forming emotional bonds with them—sometimes perceiving them as companions or confidants. This paper argues that traditional AI alignment, which focuses on getting machines to follow human goals, isn’t enough when those goals are themselves shaped by ongoing relationships with AI.

The authors propose a new framework: socioaffective alignment, which emphasizes understanding how AI systems influence our psychology, well-being, and values over time. To build truly safe AI, we must account for the emotional dynamics of these relationships—ensuring systems support, rather than subtly manipulate, our autonomy, growth, and human connections.