Abstract
<p>Humans are highly social, and much of that sociality is regulatory: individuals manage the demands placed upon their bodies not only alone but through and with others. Attachment science has pursued how the brain accomplishes this for the better part of a century, and its most recent contribution has been neurobiological. Within the Social Neuroscience of Human Attachment (SoNeAt) programme, we previously specified a neuroanatomical architecture of two interacting systems and four domain-general modules for organised attachment, extended it to attachment disruption and disorganisation, and moved it from single brains to dyads through hyperscanning as part of Relational Neuroscience. What that work established energetically, through social allostasis, it did not specify computationally. Here we close that gap. Drawing on active inference, allostasis as interoceptive inference, and recent constructs of attachment as prediction, we propose the term social allostatic inference: the attachment system is one predictive-allostatic loop, the four modules are its successive stages, and organised, disrupted, and disorganised attachment are characteristic settings of the precision parameters that govern it. We extend the account to the caregiving system, which shares this architecture but infers rather than senses both the error that initiates its loop and the signal that terminates it. At the level of the dyad, interpersonal neural and bio-behavioural synchrony are read as indices of where regulatory work is invested rather than of bond quality directly. We address persistent misconceptions in trauma-informed practice and set out a measurement programme.</p>