Notice something right now: the sound in the room around you, or the absence of it. The slight weight of whatever you’re holding, or the surface your arms rest on. A thought that was moving through your mind a moment ago and surfaced again just now when you were prompted to look for it.

You didn’t decide to become aware of any of those things. Something selected them for you—passed them through a filter you’ve been using your entire life without knowing its name.

That filter is a neural network neuroscientists call the salience network. This article introduces it not as an abstract concept, but as something already active in you—a system with a name, a location, and a mechanism that quietly shapes every moment of conscious attention you have.

You've Always Used It—You Just Didn't Know What to Call It

The salience network is not new. It predates language, sleep cycles, and the reflex of reaching for a phone in the first seconds of waking. What is relatively new is our ability to map where it lives in the brain and how it makes its decisions.

Neuroscientists began identifying its structure in the mid-2000s using functional magnetic resonance imaging (fMRI), which tracks blood flow as a proxy for neural activity. By studying which regions activated together during a wide range of experiences—and quieted together when those experiences ended—they identified a network with two primary hubs: the anterior insula (AI) and the dorsal anterior cingulate cortex (dACC).

These regions are physically distinct. The anterior insula sits buried within a fold of the cerebral cortex called the lateral sulcus, roughly behind the temple on each side. The dorsal anterior cingulate cortex arcs along the inner wall of the frontal lobe near the top of the brain, curving just above the corpus callosum. What links them is not their proximity but their shared function: together, they assess which signals—from the thousands arriving every moment from body, environment, and memory—are significant enough to reach conscious awareness.

A 2026 scoping review published in the Journal of Integrative Neuroscience by Jayashankar and colleagues described the salience network as exerting “context-dependent modulatory influences” on the brain’s other large-scale networks, operating through oscillatory coupling and directed connectivity rather than simple on-off switching (Jayashankar et al., J Integr Neurosci, 2026). The network is, in short, a regulator—not a switch.

The Two Regions Behind the Filter

Each hub brings a different competency to the network.

The anterior insula specializes in interoception: the brain’s continuous monitoring of internal body state. Heartbeat, breathing rhythm, gut pressure, skin temperature, hunger, fatigue. These signals don’t arrive as data points; they arrive as something closer to tone. The anterior insula integrates them into a continuously updated read of how the body currently is, and that read feeds directly into what the brain treats as significant.

This is part of why emotional experiences feel physical. The anterior insula isn’t generating feelings from nowhere—it’s translating body signals into a quality of experience the rest of the brain can act on.

A 2024 preregistered study published in eLife by Teed and colleagues examined this function in 92 participants (46 with anxiety, depression, or eating disorders; 46 healthy controls). Using two interoceptive tasks—one where participants directed attention toward their own cardiorespiratory signals, one where physiological changes were induced by intravenous infusion—the researchers found that the dysgranular mid-insula serves as a key cortical interface where top-down attention meets bottom-up visceral input. The anterior insula specifically activated during anticipatory uncertainty—the moment just before a body signal arrived—and this pattern was disrupted in the clinical group (Teed et al., eLife, 2024).

The dorsal anterior cingulate cortex brings a different function: conflict monitoring. The dACC responds to mismatches—when something unexpected occurs, when competing demands clash, when an ongoing action needs to change. If the anterior insula tracks how the body is, the dACC tracks when something doesn’t fit: when the current situation warrants a shift in what the brain is paying attention to.

A 2026 study published in PLOS Biology by Qian, Yue, Ng and colleagues tracked these two hubs longitudinally in older adults and found that when structural-functional integration within this anterior subnetwork declined over time, participants showed greater decline in cognitive flexibility, measured by the Trail Making Test. The relationship ran in the opposite direction for the default mode network, suggesting the two systems balance each other through the salience network’s influence (Qian et al., PLoS Biol, 2026).

The Network That Links Inward and Outward

Your brain runs at least two other large-scale networks continuously alongside the salience network. The default mode network (DMN) is active during internally directed processing: mind-wandering, self-reflection, memory replay, imagining future scenarios. The central executive network (CEN) is engaged during deliberate, externally directed work: planning, problem-solving, sustained focus on a task.

For most of waking life, you move between these modes. Something outside demands your attention, and deliberate focus needs to engage quickly. An internal state—a building worry, a body signal that won’t settle—becomes urgent enough to interrupt the task at hand. The salience network regulates these transitions.

The triple network model—the framework describing SN, DMN, and CEN as a coordinated system—has been central to cognitive neuroscience for over a decade. The 2026 scoping review by Jayashankar and colleagues examined this model in detail and found that the salience network is not a simple toggle between the other two. It exerts context-dependent influences: suppressing the DMN during demanding external tasks, but also capable of maintaining DMN engagement when internal processing is genuinely relevant to what’s happening. The review also noted that contemporary research now recognizes the DMN as dynamically engaged in both internal and task-related processing—not solely a resting or task-negative system (Jayashankar et al., J Integr Neurosci, 2026).

This reframes what it means, at the neural level, when attention wanders. When focus drifts from a task, the DMN hasn’t simply overpowered the CEN. The salience network has reassigned priority. Whatever pulled attention away was, by the brain’s own assessment, more significant at that moment than the task.

When the Body Sends a Signal, the Brain Decides How Loud It Is

Salience is not only about which external events get noticed—it is also about how loudly the brain registers its own internal state.

A 2025 study published in Molecular Psychiatry by Ince, Harrison, and colleagues at the University of Melbourne used 7-Tesla ultra-high-field fMRI and dynamic causal modelling to examine salience network function during emotional processing. The study compared 37 adults with mood and anxiety disorders to 37 healthy controls during an emotional oddball paradigm: presentations of negative stimuli interspersed with neutral ones.

In the healthy control group, the researchers identified an inhibitory signal running from the periaqueductal gray (PAG)—a midbrain structure involved in threat processing—to the anterior insula. This signal functioned as a modulator on the insula’s salience response: when a negative event occurred, the PAG dampened the anterior insula’s reactivity, preventing an outsized urgency assignment. In the group with mood and anxiety disorders, this inhibitory effect was absent. The posterior probability was 1.00, and leave-one-out cross-validation confirmed that the presence or absence of this single inhibitory signal was sufficient to predict diagnostic status (Ince et al., Mol Psychiatry, 2025).

The researchers interpreted the finding as a mechanism underlying maladaptive salience attribution: the persistent experience that something is urgent, without an external signal to account for it. When the inhibitory modulation is absent, signals that a calibrated salience network would process and release continue to be flagged as significant.

A 2026 study in Molecular Psychiatry by Brandman and colleagues extended this picture into naturalistic conditions, testing whether a specific EEG pattern reflecting amygdala activity tracked salience network dynamics during everyday emotional experience. Working with 71 healthy participants who watched film clips and listened to music while undergoing simultaneous EEG and fMRI recording, they found that the amygdala EEG pattern correlated with activity in the anterior insula and dorsal anterior cingulate cortex during emotional engagement. It also correlated with how strongly participants reported reacting to the stimuli. A control EEG pattern from a different brain region showed no such relationship (Brandman et al., Mol Psychiatry, 2026).

The participants were watching films and listening to music. The salience network was doing, in those conditions, exactly what laboratory studies describe it doing under controlled ones: continuously tracking which signals mattered more, and how much they mattered.

What This Research Does Not Establish

The studies reviewed here are heterogeneous in method, sample size, and population. The Melbourne study used 74 participants total; the eLife study used 92 in a preregistered design but at a single time point. Longitudinal data—from the PLoS Biology aging study—are more informative than cross-sectional snapshots, but the sample sizes remain modest by epidemiological standards.

The triple network model itself continues to evolve. The 2026 Journal of Integrative Neuroscience review noted that network boundaries shift depending on atlas choice, imaging modality, and task design, and that different methodological approaches “converge on the view that networks interact through flexible, frequency-specific dynamics, but diverge in their anatomical assignments.” This is a productive scientific disagreement, not a refutation of the framework—but it is a genuine limitation of any account that treats the salience network as a fixed anatomical structure with clean, permanent borders.

Population-level findings describe tendencies and mechanisms. They cannot predict how any individual brain will assign salience under any given condition. The same signal—a rapid heartbeat, a sudden sound, a body sensation that won’t resolve—will be processed differently depending on current arousal state, prior sleep, accumulated stress, and individual differences in the very networks this article describes. These studies say what the system does on average. They do not say what it is doing for any specific person at any specific moment.