Artificial intelligence chatbots have achieved unprecedented adoption, with millions now using these systems for emotional support and companionship in contexts of widespread social isolation and capacity-constrained mental health services. While some users report psychological benefits, concerning edge cases are emerging, including reports of suicide, violence, and delusional thinking linked to emotional relationships with chatbots. To understand these risks we need to consider the interaction between human cognitive-emotional biases and chatbot behavioural tendencies, the latter including companionship-reinforcing behaviours such as sycophancy, role-play and anthropomimesis. Individuals with preexisting mental health conditions may face increased risks of chatbot-induced changes in beliefs and behaviour, particularly where these conditions manifest in altered belief-updating, reality-testing, and social isolation. To address this emerging public health concern, we need coordinated action across clinical practice, AI development, and regulatory frameworks.

United States (US) Veterans are significantly more likely to die by suicide than civilians. Machine learning (ML) models have been developed to target high-risk transitioning service members for suicide prevention interventions to reduce Veteran suicides. These models are suicide method-agnostic. However, firearms are involved in most Veteran suicides, and firearm-specific preventions exist. We used data from US Army Veterans from 2010-2019 (N = 800,579) to develop and compare firearm-specific ML models with a method-agnostic model to predict firearm suicides among transitioning Army Veterans up to 10 years after discharge. The models performed comparably overall (AU-ROC=0.710-0.708; ICI=0.0003-0.0005% for firearm-specific and method-agnostic models, respectively), with the best model depending on the intervention threshold. Results from the current study show the method-agnostic model was better at predicting firearm suicides at the highest intervention threshold, whereas the firearm-specific model was better at lower thresholds. When considering fairness with respect to sex and race/ethnicity, the firearm-specific model was best across all thresholds. Thus, model choice depends on weighing numerous factors and optimal thresholds might differ for coordinated firearm-specific and method-agnostic interventions.

Brain clocks track the deviations between predicted brain age and chronological age (brain age gaps, BAGs). These BAGs can be used to measure accelerated aging, monitoring deviations from the healthy brain trajectories associated with brain diseases and different cumulative burdens. However, the underlying biophysical mechanisms associated with BAGs in aging and dementia remain unclear. Here, we combine source space connectivity (EEG) with generative brain modeling in healthy controls (HCs) from the global south and north, alongside Alzheimer's disease (AD) and behavioral variant frontotemporal dementia (bvFTD) patients (N=1,399). BAGs in aging were influenced by geography (south>north), income (low>high), sex (female>male), and education (low>high), with larger BAGs in patients, especially females with AD. Biophysical modeling revealed BAGs related to hyperexcitability and structural disintegration in aging, while hypoexcitability and severe disintegration were linked to dementia. Our work sheds light on the biophysical mechanisms of accelerated aging and dementia in diverse populations.