Abstract: Safety learning refers to an adaptive psychological mechanism through which individuals identify safety signals that predict the absence of threat, thereby inhibiting fear responses and facilitating effective adaptation to complex environments. Impairments in this mechanism have been closely linked to a wide range of anxiety and related disorders. However, existing research has predominantly interpreted the pathogenesis of anxiety from the perspectives of fear learning or fear extinction, which has proven insufficient to fully account for core clinical features such as difficulties in safety discrimination, diminished positive emotional experience, and failures in fear inhibition. In light of these limitations, the present article systematically reviews evidence from animal models, human behavioral studies, and neuroimaging research, and proposes a three-stage neural framework of safety learning—comprising safety perception and evaluation, safety acquisition, and safety expression—to more comprehensively characterize the dynamic processes of safety learning and their abnormalities in anxiety disorders.
According to this framework, safety learning can be divided into three stages: (1) Safety perception and evaluation: At this stage, individuals detect external cues and rapidly evaluate their safety value. This process primarily relies on functional networks involving the thalamus, the sensory cortices, the insula, and the amygdala, which jointly support the early analysis of whether a stimulus signals safety. (2) Safety acquisition comprises two interrelated sub-processes: The first is the formation and maintenance of safety associations, whereby stable links between conditioned stimuli and non-threatening outcomes are established through prediction-error-driven learning. The second is the acquisition of positive affect, whereby repeated exposure to safe outcomes confers positive emotional value to safety cues. This stage is supported by midbrain-striatal dopaminergic circuits that facilitate both safety association learning and positive emotional experience, while the long-term consolidation of safety associations depends on coordinated interactions between the hippocampus and the prefrontal cortex, providing a foundation for the subsequent retrieval and utilization of safety memories.(3) Safety expression, which refers to the effective use of learned safety signals to inhibit fear responses. The neural mechanisms underlying this stage involve interactions among limbic regions (such as the hippocampus and amygdala), the prefrontal cortex, and temporal cortical areas. These regions jointly support the context-dependent retrieval of safety memories, thereby enabling the suppression of fear-related behavior.
Building on this framework, the article further delineates stage-specific neural deficits in individuals with anxiety and related disorders. During the perception and evaluation stage, these individuals exhibit exaggerated threat perception and insufficient top-down cognitive regulation. During the acquisition stage, they show reduced efficiency in prediction error coding and attenuated positive emotional experience associated with safety signals. During the expression stage, impairments are evident in the retrieval of safety memories and in the inhibition of fear responses. Each of these behavioral abnormalities is accompanied by corresponding dysfunctions in specific brain regions and neural circuits.
Finally, the article highlights several directions for future research and clinical application. First, future studies should strengthen the use of conditioned inhibition paradigms in investigations of safety learning in anxious populations. To enhance ecological validity and translational relevance, researchers are encouraged to incorporate naturalistic stimuli with semantic or social attributes and to combine multimodal neural measurements with multivariate analytical approaches to systematically examine neural representations of safety learning and recall in anxiety. Second, given the pervasive deficit in positive emotional experience during safety learning among anxious individuals, future interventions should place greater emphasis on enhancing the encoding of safety value and positive emotional experience. Third, at the level of clinical translation, the proposed framework underscores the dynamic, multi-stage nature of safety learning, suggesting that intervention strategies should be informed by underlying neural mechanisms and developmental stages, and implemented in a staged and targeted manner to achieve more precise and effective treatment outcomes.

Key words: safety learning, anxiety, neural mechanisms