Can We Control Anxiety Without Sedating the Mind?

Almost nobody who suffers from anxiety wants to feel nothing. They want to keep the part of the system that notices a swerving car and stop the part that replays an awkward sentence from three weeks ago. The wish is for selectivity: less false alarm, no loss of the capacity to be alerted by a real one.

That wish is where the technology gets hard. The brain’s threat machinery is not a single circuit that can be turned down. It is a network that decides what is salient, prepares the body for action, and learns from outcomes — and the same machinery that produces crippling worry also produces vigilance, quick reactions, and the ability to reconsider a decision when something feels off. Sedation works because it damps the whole thing. The question is whether anything else can.

As in the rest of this series, three categories have to be kept apart. Some findings are demonstrated in controlled human studies. Some are plausible engineering that follows from those findings but has not been built or proven. Some are speculation dressed as a product roadmap. The temptation to collapse them is especially strong here, because anxiety is common, treatment is imperfect, and the market for relief is enormous.

What anxiety is made of

Anxiety in the clinical sense is not a feeling that sits in one place. It involves a network centered on the amygdala and its connections to the hippocampus, insula, and medial prefrontal cortex, plus the autonomic nervous system that translates appraisal into heart rate, breathing, sweating, and gut sensation, plus the cortical systems that appraise and sometimes override all of it. These components can dissociate. A person can show a large skin-conductance response to a threat cue while reporting no fear, or report intense fear with a flat physiological response. That dissociation is one reason the field keeps arguing about what a treatment is supposed to change.

It is also why the aspiration of “reducing anxiety without sedating the mind” is not one goal but several. Reduce the frequency of false alarms. Reduce the intensity of the body’s response once an alarm fires. Preserve the ability to detect genuine danger. Preserve working memory, reaction time, and the capacity to act. A drug that lowers arousal does all of these at once, in the same direction, including the ones a person wants to keep. Anything more selective has to target a narrower piece.

The clinical baseline matters before any technology enters the picture. Standard treatments — selective serotonin reuptake inhibitors, cognitive behavioral therapy with exposure, and their combination — work for many people and fail for a substantial minority, and they take weeks to work when they work at all. Benzodiazepines act quickly but carry tolerance, dependence, and cognitive dulling. That gap is what makes the research below worth taking seriously, and it is also what makes it easy to overstate.

The breathing channel

The cheapest intervention for anxiety is also the oldest: change how you breathe. There is a real mechanism underneath the folk practice, though it has been demonstrated mostly in animals.

Kevin Yackle and colleagues identified a small population of neurons in the preBötzinger complex, the brainstem region that generates the respiratory rhythm, that project to and positively regulate noradrenergic neurons in the locus coeruleus (Yackle et al., 2017). When the researchers ablated roughly 175 of these neurons in adult mice, breathing remained intact, but the animals spent more time in calm behaviors and less in aroused states. The locus coeruleus is the same structure implicated in alertness and panic, with diffuse projections across the forebrain.

The finding is a mechanism, not a therapy. It shows that the rhythm of breathing is wired into the arousal system rather than merely correlated with it, which gives slow breathing a plausible causal route to a calmer state. It does not show that any breathing protocol treats an anxiety disorder, and the study was done in mice with a targeted cell ablation. Practically, the interesting part is the direction of the coupling. Breathing appears to be one of the few autonomic processes a person can control directly, and it sits upstream of a system that is otherwise hard to reach by intention.

Biofeedback: measurable physiology, uncertain change

If breathing is a manual lever, heart-rate variability biofeedback is the instrumented version. A person breathes at a rate that maximizes the oscillation in heart period, watches a live display of that oscillation, and practices until the pattern becomes easier to produce.

The strongest summary evidence is encouraging. A meta-analysis of 24 studies and 484 participants found a large pre-post reduction in self-reported stress and anxiety, with a Hedges’ g of 0.81 within groups and 0.83 when biofeedback was compared against a control condition (Goessl et al., 2017). The authors noted that the effect was not moderated by study year, risk of bias, the proportion of female participants, the number of sessions, or whether participants had a diagnosed anxiety disorder.

That last point cuts both ways. It is a striking result, and the intervention is cheap, non-invasive, and self-administered. It is also built almost entirely on self-reported outcomes in small trials where participants knew they were receiving a treatment and could not be blinded to it. Self-report is a legitimate outcome — how anxious a person feels is the thing they came to change — but it is also the outcome most vulnerable to expectancy, especially when the treatment involves visible effort and visible progress. A meta-analysis cannot repair a design flaw that runs through all the studies it pools.

A more recent randomized, active-controlled trial makes the gap concrete. Ryuji Saito and colleagues assigned 55 participants with elevated anxiety to heart-rate variability biofeedback or an active control and measured both physiology and brain response with EEG over ten sessions (Saito et al., 2024). The biofeedback group increased its resting vagal tone relative to controls. It did not show corresponding group differences on the anxiety questionnaire, on attentional bias to threat, or on the event-related and time-frequency EEG measures.

This is the single most instructive result in the literature on autonomic training. The physiological target moved. The person-facing outcome did not, at least not detectably, in a study designed to detect it. That does not invalidate the meta-analytic signal, and it does not mean vagal tone is irrelevant. It means the chain from “changed a bodily signal” to “changed how someone feels and functions” is longer than it looks, and that researchers often measure the proxy because it is easy and the outcome because they must.

Ultrasound on the amygdala

The most direct non-invasive approach to the threat network is low-intensity transcranial focused ultrasound, which can deliver acoustic energy to a structure several centimeters deep without surgery. Because the amygdala sits well below the reach of conventional transcranial magnetic stimulation, this is the first method that can target it from outside the skull.

Two human studies point in encouraging directions. In a double-blind, sham-controlled study of 30 healthy adults, Tina Chou and colleagues applied active or sham ultrasound to the left amygdala and then measured brain response during a fear-inducing task (Chou et al., 2024). Compared with sham, active stimulation reduced blood-oxygen-level-dependent activation in the amygdala, hippocampus, and dorsal anterior cingulate during the fear task. It also changed resting connectivity: less coupling between the amygdala and the insula and hippocampus, more between the amygdala and the ventromedial prefrontal cortex, a pattern often associated with better top-down regulation. The size of the drop in amygdala activation correlated with the size of the drop in self-reported anxiety — about 0.62 across individuals. There was no group difference in skin-conductance responses, which is a reminder that different measures of “anxiety” can move independently.

The second study extends this toward patients. In a target-engagement experiment followed by an unblinded single-arm trial, Bryan Barksdale, Gregory Fonzo, and colleagues administered ultrasound to the left amygdala in 29 people with mood, anxiety, and trauma-related disorders and 23 healthy comparison subjects (Barksdale et al., 2025). Active stimulation reduced amygdala signal relative to sham, replicating the earlier target engagement. The same patients then received daily sessions for three weeks. Treatment was well tolerated with no serious adverse events, and the primary clinical outcome improved with a large effect size. The authors framed the result correctly: it motivates double-blind randomized trials, and it does not establish efficacy. An unblinded single-arm trial cannot separate the intervention from expectation, regression to the mean, or the natural course of illness.

And then there is the study that keeps this literature honest. Bianca Hoang-Dang and colleagues tested amygdala ultrasound in 21 healthy older adults using a within-participant crossover design with an active control region (Hoang-Dang et al., 2024). Targeting the amygdala increased self-reported arousal in response to negative images and increased heart rate. Targeting the entorhinal cortex did not. State anxiety did not change.

That is the mirror image of the result above, in a different population with different parameters and a different target within the amygdala. The honest reading is not that one team is wrong. It is that the direction of the effect depends on the acoustic parameters, the precise target, the person’s age and clinical status, and the timing of measurement. A tool that can calm a fear network in one configuration and heighten reactivity in another is a research instrument, not a dial, and the discrepancy is exactly why double-blind sham control is not optional in this field.

Closed loops and clinical reality

Most stimulation is open-loop: a fixed dose on a fixed schedule, with no measurement of whether it is needed. The alternative is to detect a symptom-relevant signal and intervene only when it appears.

The proof of concept comes from obsessive-compulsive disorder rather than anxiety as such, but the logic transfers. In a first-in-human case, Young-Hoon Nho and colleagues implanted a responsive neurostimulation device in a patient with severe, treatment-resistant OCD who also had epilepsy (Nho et al., 2024). Over months of ambulatory recording, the team identified a signal in the ventral striatum — low-frequency oscillatory power in the nucleus accumbens and ventral pallidum — that tracked the patient’s obsessions and provoked distress. They configured the device to deliver brief bursts of stimulation when that signal crossed a threshold. The authors reported rapid, robust, and durable improvement in obsessions and compulsions.

Two features distinguish this from a general anxiety treatment. First, it required a surgically implanted device that records from deep brain structures, which is only conceivable for people with severe, refractory illness. Second, it worked because OCD has an episodic, symptom-linked physiology that fluctuates on a timescale a device can detect. Anxiety, as most people experience it, is less like a storm that arrives and more like a background climate. Detecting the moment to intervene is the central unsolved problem.

The regulatory frame is worth stating plainly. Deep brain stimulation of the ventral anterior limb of the internal capsule has been available for treatment-resistant OCD in the United States since 2009, but under a Humanitarian Device Exemption rather than full premarket approval (FDA, HDE H050003). The approved indication is narrow: bilateral stimulation as an adjunct to medication and an alternative to anterior capsulotomy, in adults who have failed at least three selective serotonin reuptake inhibitors. Labeling for devices approved on this pathway is required to state that effectiveness for the labeled indication has not been demonstrated, and the pathway exists precisely because the condition is rare enough that conventional trials are impractical. Even the flagship invasive treatment for a severe anxiety-spectrum disorder carries that caveat. Long-term outcomes in the largest clinical cohorts are mixed, and response rates in OCD hover near half, with substantial heterogeneity in who benefits.

Treating a disorder and tuning a person

Everything above concerns people who are ill. The moment the target shifts from a disorder to ordinary human variation, several things change at once.

The first is the ratio of benefit to cost. For someone whose OCD or panic disorder has eliminated work, relationships, and sleep, a surgical implant with a real adverse-event profile is a rational trade. For someone with mild test anxiety, almost nothing is. The acceptable side-effect burden scales with the severity of the condition, and treatments that clear the bar in one population fall far below it in the other.

The second is that anxiety is informative, at least sometimes. It is not a malfunction that should be minimized to zero. Worry motivates preparation, dread of consequences restrains harmful behavior, and a low-grade sense of unease is often the first signal that a plan is wrong. The clinical literature distinguishes excessive, impairing anxiety from useful caution, and any tool aimed at the former has to avoid eating into the latter. This is the selectivity problem again, and it is not a marketing detail.

The third is that ordinary anxiety is often appropriate to a bad situation, and dampening it can hide the problem. A worker who is anxious because the workload is unsustainable, a student anxious because the material is beyond their preparation, and a patient anxious because a symptom is being dismissed are all receiving accurate signals. Suppressing the signal makes the person more comfortable and the situation worse. Any tool that treats distress as the thing to be minimized will be used this way, and the incentives of employers, schools, and insurers point in exactly that direction.

What selectivity would have to mean

Suppose a selective intervention existed. What would distinguish it from sedation?

It would reduce reactivity to threat cues while leaving performance on a demanding cognitive task intact. It would reduce the frequency of false alarms while leaving detection of real ones unchanged, which can be tested by measuring response to a genuine, clearly signaled threat alongside response to an ambiguous cue. It would show effects that persist after the intervention stops, because a treatment that works only while the device is on has not taught the person anything. Its benefits would need to be established with a congruent sham control, because this field has already produced opposite results in similar designs. Primary outcomes would have to be pre-registered, and the measures would need to include function and alertness, not just symptom scores, so that a global blunting effect could be detected and rejected. Adverse events would need explicit reporting of anxiety worsening, because the Hoang-Dang result shows that increased reactivity falls within the range of plausible outcomes.

And it would need follow-up that outlasts the enthusiasm of the trial, since the deepest risk in this domain is dependence. A person who can only tolerate a difficult meeting with a device switched on has traded one constraint for another. The relevant test is not whether anxiety falls while the intervention runs. It is whether the person can do more, unaided, six months later.

Where this leaves the question

The evidence supports a modest version of the original hope. Threat responses can be measured in real time. In controlled laboratory conditions, deep structures in the fear network can be modulated from outside the skull, with measurable changes in brain activity and, in some studies, in subjective anxiety. Autonomic training can shift the physiology of stress. In a narrow, severe clinical population, a closed-loop implant has demonstrably reduced symptoms that had resisted everything else.

The evidence does not support the larger version. Nobody has a dial for anxiety. The effects are small, parameter-sensitive, and sometimes reversed; the person-level outcomes lag the physiological ones; the strongest clinical result in the area comes from a single case, and the flagship device carries a regulatory statement that its effectiveness has not been demonstrated. Between the laboratory finding and the useful, selective, non-sedating intervention sits the same distance that separates a laboratory result from a product in every other part of this series.

The harder question is not technical. It is who decides what a person’s arousal should be set to, and in whose interest. For a patient choosing relief from a disabling illness, the answer is straightforward. For everyone else, the answer will be set by whoever is paying — and the least anxious person in the room is rarely the one with the most power over the setting.