Abstract: Large language models (LLMs) may exhibit unintended or undesirable behaviors. Recent works have concentrated on aligning LLMs to mitigate harmful outputs. Despite these efforts, some anomalies indicate that even a well-conducted alignment process can be easily circumvented, whether intentionally or accidentally. Does alignment fine-tuning yield robust effects on models, or are its impacts merely superficial? This article makes the first exploration of this phenomenon from both theoretical and empirical perspectives. Empirically, it demonstrates the elasticity of post-alignment models, i.e., the tendency to revert to the behavior distribution formed during the pre-training phase upon further fine-tuning. Leveraging compression theory, it formally deduces that fine-tuning disproportionately undermines alignment relative to pre-training, potentially by orders of magnitude. This article validates the presence of elasticity through experiments on models of varying types and scales. Specifically, it find that model performance declines rapidly before reverting to the pre-training distribution, after which the rate of decline drops significantly. Furthermore, this article reveals that elasticity positively correlates with the increased model size and the expansion of pre-training data. The research findings underscore the need to address the inherent elasticity of LLMs to mitigate their resistance to alignment.