A face that once predicted a shock can become safe. The difficult part is not noticing that the shock has stopped; it is updating the nervous system quickly enough for that new information to matter. A large brain-imaging study published on November 7 suggests that this updating process unfolds differently, on average, in people with anxiety disorders.
The finding is more subtle than a “fear center” that refuses to switch off. Across 383 adults, the researchers observed less change in brain activation over repeated safe trials in the anxiety group than in healthy controls. The difference appeared for both a former threat cue and a cue that had always been safe. It points toward a broader problem of adapting to changing information—but it does not show that the scan can diagnose anxiety, explain any one person’s symptoms, or predict whether exposure therapy will work.
Extinction adds safety; it does not erase fear
The study, published in *Translational Psychiatry*, drew on the German multicenter PROTECT-AD project. It included 280 people with panic disorder, agoraphobia, social anxiety disorder or multiple specific phobias, plus 103 healthy controls.
On the first day, participants saw neutral faces. One face, the conditioned threat cue, was followed by a mild electric shock on 60 percent of its appearances. Another face was never paired with shock. On the second day, participants entered an fMRI scanner and saw both faces repeatedly without receiving any shock. The researchers compared blood-oxygen-level-dependent activity during early and late blocks of this extinction phase.
“Extinction” can sound like deletion. In learning research, it usually means something else: the person learns a new association—this cue is now safe—that competes with the older threat association. The original memory can return in a different context or after time has passed. That distinction helped inspire an inhibitory-learning approach to exposure therapy, which emphasizes learning that feared outcomes do not occur rather than simply waiting for distress to decline.
The new experiment tested a laboratory analogue of that process, not psychotherapy itself. That boundary matters. A face-shock task in a scanner captures a controlled slice of threat learning; clinical exposure unfolds across personally meaningful situations, expectations and choices.
The difference was broad—and not a simple failure to calm down
From early to late extinction, patients showed less attenuation of activity than controls across a network that included the insular cortex, supplementary motor areas and cingulate regions. A focused analysis of the left anterior insula supported the group difference. Yet there was no significant interaction between group and cue type. The altered trajectory was not confined to the face that had predicted shock; it also appeared for the always-safe face.
That result complicates a familiar story in which anxiety is only excessive reactivity to danger. The pattern may reflect a more general difficulty adjusting neural responses when the environment changes. Participants with anxiety disorders also reported higher shock expectancy and arousal during extinction, suggesting that the new safety information had less subjective impact.
But “less attenuation” should not be translated as “the anxious brain stayed more active.” The group difference was driven largely by the early extinction window, when patients showed lower activation than controls in several regions. Because fMRI measures blood-flow-related signals rather than neural firing directly, the same early-to-late contrast can arise from different starting points. Scanner stress, anticipation and other state differences may also shape the first trials. The authors adjusted an analysis for state anxiety, but that cannot eliminate every alternative explanation.
The study therefore contributes a dynamic clue, not a single biomarker. Its strength is the unusually large, diagnostically varied sample and the same delayed protocol across multiple sites. Its most interesting message is about change over time: averaging the entire extinction phase could hide the point at which groups diverge.
What the scan cannot tell us about treatment
The authors describe reduced attenuation as a transdiagnostic feature, meaning that it appeared across several anxiety diagnoses. That is a group-level statistical pattern. It does not establish that the pattern causes anxiety, nor that every patient shows it. Generalized anxiety disorder was not represented, and combining diagnoses increases reach while obscuring disorder-specific differences.
Other limits narrow the conclusion. The analyses were not preregistered. The block-based design offers coarse timing, so trial-by-trial modeling could reveal a different learning curve. No skin-conductance measure was available to track autonomic arousal alongside self-report and fMRI. The reported group-by-time effects in subjective ratings were also less clear than the persistent overall differences between groups.
Most importantly, the experiment did not test treatment response. Exposure therapy is informed by extinction research, but the link is not one-to-one. A review of fear extinction and relapse emphasizes that learning safety and retrieving that learning later are separate challenges. This study measured learning during one laboratory session, not later recall, real-world avoidance or clinical improvement.
The practical consequence is restraint with a useful shift in focus. Anxiety research may learn more by asking how brains update across moments than by hunting for one overactive region. The new data make that question sharper: when safety becomes available, the decisive signal may be not how strongly the brain reacts, but how flexibly its response changes.
