Abstract value concepts such as "wellbeing," "safety," and "optimization" are widely used as top-level guiding principles in the design of personal AI systems. While such concepts appear ethically desirable and implementation-agnostic, their high level of abstraction introduces systematic risks of conceptual drift, scope expansion, and unintended ethical distortion. This paper analyzes how abstract value anchors function within personal AI architectures and identifies three core risk patterns: semantic expansion beyond designer intent, progressive reinterpretation through interaction, and loss of constraint through abstraction stacking. By clarifying the structural mechanisms through which abstract values destabilize ethical alignment, this study provides a conceptual risk framework for designers of personal AI systems. The aim is not to reject abstract values, but to highlight the design-level vulnerabilities they introduce when deployed without explicit scope constraints and validation mechanisms.