The research, led by Yina Ma of Beijing Normal University together with neurosurgeons at Beijing Tiantan Hospital and the Chinese PLA General Hospital, took advantage of a rare scientific opportunity. Patients with drug-resistant epilepsy sometimes undergo intracranial electroencephalography, in which electrodes are implanted directly in the brain to locate seizure origins. The team recruited pairs of these patients and had them play a real-time, iterated prisoner’s dilemma game while recording neural activity from both brains simultaneously. This setup, known as hyperscanning, allowed the researchers to observe, millisecond by millisecond, what happens inside two interacting brains as they negotiate the oldest dilemma in game theory.

The central discovery concerns something the authors call interbrain representational alignment, or IBRA. This is a measure of how similar the fine-grained spectral patterns of neural activity are between two partners at the same moment. The researchers found that this alignment emerged selectively in the amygdala, a small almond-shaped structure deep in the temporal lobe long associated with emotion, threat detection and social valuation, precisely during trials in which both players chose to cooperate. When only one person cooperated, or when both defected, the alignment did not appear. Even more strikingly, the strength of this amygdala alignment predicted, trial by trial, whether mutual cooperation would actually occur.

The alignment was not a single phenomenon but a two-stage cascade unfolding across the roughly 1.5 seconds of each decision. In an early window, beginning around 100 milliseconds after the game stimulus appeared, alignment was characterized by suppression of theta-band activity, the slow oscillations between roughly 4 and 8 hertz. This early phase appeared to reflect rapid, history-sensitive coordination: the brains seemed to be quickly taking stock of what had happened in previous rounds and settling into a shared state. In a later window, from roughly half a second onward, the alignment shifted its signature, combining suppression of low-frequency activity with increases in gamma-band power, the fast oscillations above roughly 30 hertz associated with active neural computation. This later phase appeared to support coordinated commitment, the moment when both partners lock in their cooperative intentions and prepare their actions.

What makes the amygdala’s role particularly compelling is what it was computing. The researchers built computational models of how players estimate their partner’s likely behavior, drawing on reinforcement-learning frameworks in which each player maintains a running, model-based prediction of whether the partner will cooperate on the next trial. They found that amygdala low-frequency suppression tracked this prospective estimate of partner cooperation. In other words, the amygdala was not merely reacting to the partner’s past moves; it was simulating the partner’s future decision. Theta-gamma coupling, a form of cross-frequency interaction in which the phase of slow oscillations modulates the amplitude of fast ones, then linked this simulated estimate of the partner to the player’s own decision process, completing what the authors describe as a simulation-to-decision cascade.

This finding resonates with earlier work in nonhuman primates. Studies by other groups have shown that individual neurons in the macaque amygdala appear to simulate the decision processes of social partners, and that other amygdala neurons predict an opponent’s behavior during cooperative exchanges. The new study extends this simulation hypothesis to humans with unprecedented temporal precision, and adds a crucial twist: the simulation does not stay contained within one brain. When two people successfully cooperate, their simulated representations become spectrally similar across skulls, a kind of neural handshake that no single-brain study could ever detect.

The study also revealed that not all cooperators are alike. By measuring personality traits such as social value orientation and tolerance of uncertainty, the researchers distinguished two behavioral types. Conditional cooperators are strategic players who cooperate only when they believe their partner will reciprocate; they leaned heavily on the amygdala’s partner-tracking machinery, adjusting their neural alignment to the flow of the game. Unconditional cooperators, by contrast, cooperate almost regardless of history. These individuals appeared to rely on a dispositional tolerance of uncertainty, an amygdala-related trait that allowed them to transition into anticipatory neural alignment even before there was any evidence that the partner would reciprocate. In effect, unconditional cooperators extend trust first and let the interbrain alignment follow.

The methodological rigor behind these conclusions is considerable. Because simultaneous intracranial recordings from two people are extraordinarily rare, the team confronted a host of potential confounds. Could the alignment simply reflect similar timing of decisions, shared sensory input, or volume conduction between electrodes? The authors controlled for dyadic decision time and decision-time similarity using propensity-score matching, re-referenced signals to white matter to eliminate conductive artifacts, excluded repeated electrode pairs, replicated the effect with linear mixed-effects models treating dyads, participants and channel pairs as nested random effects, and showed that no alignment appeared during the pre-stimulus fixation period when nothing social was happening. Time-lagged analyses even revealed directionality: the neural spectra of the partner with the stronger predictive signal tended to lead those of the other, and the faster decision-maker’s brain led the slower one’s, suggesting a genuine informational flow rather than a coincidental echo.

Control analyses in the temporoparietal junction, a cortical region implicated in real-time social interaction, showed alignment in both cooperative and non-cooperative conditions, indicating that some degree of interbrain coupling is a general feature of live interaction. The amygdala’s contribution is different in kind: it is the selective, outcome-predictive alignment that appears only when two independent minds converge on mutual trust. A linear classifier trained on amygdala alignment strength could decode, from held-out dyads, whether a given trial would end in mutual cooperation, in both the early and late time windows.

The broader implications reach well beyond the laboratory. Social dilemmas, from climate agreements to everyday trust, hinge on the same uncertainty the patients faced: will the other side cooperate? This study suggests that the answer is computed not in isolation but through a biological dialogue, in which the amygdala simulates the partner, couples that simulation to one’s own choice, and, when cooperation succeeds, locks the two brains into a shared spectral rhythm. Cooperation, on this view, is less a decision than an alignment, a temporary merging of two predictive machines into one coordinated system. Understanding that mechanism may eventually illuminate why some people find trust easy, why others guard it carefully, and how the neural interdependency beneath every handshake, contract and alliance actually works.

Subject of Research: Amygdala-mediated interbrain neural alignment and simulation-to-decision mechanisms underlying mutual cooperation in a real-time prisoner's dilemma

Article Title: An amygdala-mediated simulation-to-decision cascade aligns interbrain dynamics to solve social dilemmas

Article References: Ni, J., Meng, F., Zhang, C., Tan, H., Zeng, X., Yang, J., Li, H., Qi, Y., Yan, Y., Zhang, Y., Zhang, K., & Ma, Y. (2026). An amygdala-mediated simulation-to-decision cascade aligns interbrain dynamics to solve social dilemmas. Nature Neuroscience. https://doi.org/10.1038/s41593-026-02473-z

Image Credits: AI Generated

DOI: 10.1038/s41593-026-02473-z

Keywords: amygdala, cooperation, interbrain synchrony, intracranial EEG, prisoner's dilemma, social decision-making, theta-gamma coupling, hyperscanning, neuroscience, trust, simulation, Nature Neuroscience