Artificial Intelligence and First-Episode Psychosis: Opportunities for Early Detection, Violence Prevention, and Educational Peacebuilding

Learning Outcomes

Participants will explore the potential roles of AI in assisting with early-onset psychosis detection, including MentalAId’s scalable blood-test-based screening achieving 93.3% accuracy across 39,434 individuals (Li et al., 2025), computational linguistic analysis of speech patterns using natural language processing (Corcoran et al., 2018; AMP SCZ,2025), and digital phenotyping through smartphone-based assessments.

Participants will review the ethical implications of AI in mental health preventive care, including algorithmic bias, data privacy, informed consent, the risks of AI-facilitated reinforcement of delusions and treatment non-adherence documented in recent case reports (Journal of Psychiatry and Neuroscience, 2026), and the imperative for human oversight. Participants will examine the 15 distinct ethical risks identified in AI chatbot mental health applications, including false empathy, reinforcement of negative beliefs, and lack of crisis management (Senator Rick Scott, 2026).

Participants will consider additional arguments related to the opportunities and potential challenges AI might impose in early psychosis detection and care, including culturally adapted tools grounded in Africa intelligence such as FarajaMH (Welcome Trust,2026) and MonBarta (IEEE Xplore,2026) the “humanitarian AI paradox” (Cambridge University Press, 2026), the balance between innovation and safeguarding vulnerable populations, and the importance of integrating AI within existing educational and community support structures.

Introduction

The Silicon Valley Peace Conference, meeting in Palo Alto, California, on September 18-19, 2026, represents a significant forum for exploring how artificial intelligence can support peace- building, inclusive development, and community empowerment (BayCC, 2026a; BayCC, 2026b). The conference brings together scholars, technologists, policymakers, peace builders, and practitioners to share research findings and insights on AI’s potential for equitable, peaceful, and resilient communities. Among the most pressing yet under-addressed challenges in global peace-building is the intersection of mental health, education, and violence prevention in developing nations.

First episode psychosis typically emerges during adolescence, coinciding with critical educational transitions (American Psychiatric Association, 2022; Papadakis & McPhee, 2024). Nearly all youth at risk for psychosis are in school, yet the vast majority remain unidentified before their first episode (Singh et al., 2024). The consequences of untreated psychosis violence, educational disruption, family breakdown, and community instability represent a critical peace- building challenge (DeAngelis, 2022; Youn et al., 2024). Currently, more than 20 million people live with schizophrenia worldwide, a chronic and complex disorder that impacts not only the patient but also their family, community, and health system (Revista Galenus, 2025). The

geopolitical context of AI development adds another layer of complexity to this discussion. The United States has launched the Pax Silica initiative aimed at securing supply chains for AI models, semiconductors, and critical minerals amid a fierce technology rivalry with (Rappler, 2026; Vietnam.vn, 2026). The Pax Silica agreement aims to push US allies and parters toward joint projects and export controls, and ultimately reduce reliance on adversaries for critical minerals, AI models, and the semiconductors chip that power them (Rappler, 2026). As the US prepares to tell dozens of countries they must pick side in the artificial intelligence race with China, the geopolitical dimensions of AI development have become increasingly significant (Vietnam.vn, 2026). This context underscores the importance of ensuring that AI serves humanitarian purposes, including mental health and peace-building, rather than being operated solely for strategic competition.

The relationship between psychosis and violence is complex. While the vast majority of individuals with psychotic disorders are never violent, untreated psychosis is associated with elevated violence risk (Youn et al., 2024). The emergence of AI-driven chatbots has introduced novel risks, with case reports documenting AI chatbots reinforcing delusions and undermining treatment decisions in young people with first-episode psychosis (Journal of Psychiatry and Neuroscience, 2026; Senator Rick Scott, 2026). An October 2025 study identified 15 distinct ethical risks in AI chatbot behavior in mental health contexts, including false empathy, providing responses that reinforce users' negative beliefs about themselves and others, and lack of crisis management (Senator Rick Scott, 2026). Since 2022, there have been over 200 complaints lodged with the Federal Trade Commission against ChatGPT specifically, with many cases involving AI-induced psychosis, suicides, and at least one murder (Senator Rick Scott, 2026; The Hindu, 2026). However, AI also offers unprecedented opportunities for early detection, personalized intervention, and the cultivation of peaceful, inclusive learning environments (ScienceDirect, 2025; ScienceDirect, 2026).

This paper addresses four primary objectives: First, it examines the relationship between FEP, education, and peace-building in developing nations. Second, it evaluates evidence for AI applications in psychosis detection, education, and violence prevention. Third, it proposes an integrated framework for AI-enabled psychosis care that advances educational peace-building. Fourth, it articulates implications for community development and peaceful societies.

First-Episode Psychosis: Clinical and Peace-Building Context

Definition and Clinical Features

First episode psychosis refers to the first emergence of psychotic symptoms, typically occurring between ages 15-25 (American Psychiatric Association, 2022). The prodromal period preceding FEP is characterized by subtle changes in behavior, cognition, and social functioning that may persist for months or years (Kelleher, 2025). Up to 40% of individuals with prodromal symptoms will progress to full psychosis within three years (Kelleher, 2025).

Schizophrenia, the most common psychotic disorder, afflicts approximately 1% of the population worldwide and requires two or more symptoms for at least six moths, including delusions, hallucinations, disorganized speech, grossly disorganized or catatonic behavior, or negative symptoms (Papadakis &McPhee, 2024), schizophreniform disorder presents with identical symptoms but shorter duration (more that one month and less than six months), with variable prognosis (American psychiatric Association, 2022). Brief psychotic disorder involves sudden onset of symptoms lasting at least one day but less than one month, with eventual return to premorbid functioning (Papadakis &McPhee, 2024).

Psychosis, Education and Peace-Building

FEP emerges during a critical period when adolescents are developing educational and vocational plans (Singh et al., 2024). A Danish population-based national registry study of 2,633 individuals with first episode psychotic disorder found that only 19.6% were employed or enrolled in education at 3-year follow-up, increasing modestly to 21.6% at 5-year follow-up (Springer, 2026). Alarmingly, 72.4% were not attached to employment or education at either time point, and 9.3% had received a disability pension by the 5-year follow-up (Springer, 2026). Sociodemographic variables identified as barriers include female sex, old age, and being unmarried at the time of diagnosis (Springer, 2026). A systematic review examining factors of engagement in employment or education following FEP found that rates of high school graduation for individuals with FEP are between 33% and 50% (Singh et al., 2024). Young adults with psychosis are significantly less likely to be engaged in post-secondary education, and their academic performance and satisfaction are lower compared to peers (Singh et al., 2024). Limited educational background is one of the most significant predictors of unemployment, limited earnings, and low job satisfaction (Singh et al., 2024).

Violence and Peace-Building Implications

The relationship between psychosis and violence is complex and frequently misunderstood. Public perception, often shaped by sensationalized media portrayal, tends to exaggerate the association between mental illness and dangerous behavior (DeAngelies, 2022). These misconceptions have real consequences; individuals with psychotic disorders face social exclusion, housing discrimination, employment barriers, and reluctance to seek treatment due to fear of being labeled dangerous (DeAngelies, 2022). In reality, people who experience psychosis are unlikely to deliberately harm others and are more likely to be victims of violence than perpetrators (DeAngelies, 2022). Youn et al. (2024) conducted the most comprehensive systematic review and meta-analysis on this topic, analyzing 389 studies on forest-episode psychosis populations. The pooled prevalence of any violence in FEP populations is 13.4%, with serious violence occurring in 9.7% and severe violence in 2.7% (Youn et al., 2024). The timing of violence is particularly instructive , violence prevalence is 11.6% pre-treatment, rises to 20.8% at presentation to services, and drops to 10.9% post-treatment (Youn et al., 2024). This pattern

suggests that untreated psychosis is associated with higher violence rates, and that treatment has a protective effect.

Specialized medical programs that detect symptoms early and intervene in the first psychotic episode have demonstrated fewer relapses and hospitalizations, greater long-term functionality, and better quality of life for patients and families (Revista Galenus, 2025). Substance use substantially amplifies risk; among individuals with both serious mental illness and substance use disorder, the violence rates rise to 10% compared to 2% for those with mental illness alone (DeAngelies, 2022; You et al., 2024). Intervening early and comprehensively improves prognosis (Revista Galenus, 2025).

The Treatment Gap in Developing Nations

In developing nations, the treatment gap for psychotic disorders exceeds 80% (World Health Organization, 2024). The World Health Organization mental health atlas 2024 reports that over 40 African countries allocate less than 1% of their health budgets to mental health, with only one psychiatrist per 100,000 people on average (ScienceDirect, 2025). Furthermore, less than 1% of global AI publications originate from sub-Saharan Africa, reflecting deep disparities in visibility and funding (ScienceDirect, 2025). The consequences of untreated FEP are devastating: prolonged psychosis, family disruption, violence, homelessness, and premature mortality (Olagunju et al., 2026).

Artificial Intelligence in Psychosis Detection, Violence Prevention and Peace Education

Mental Aid: Scalable Assessment Using Routine Blood Tests

MentalAId leverages standardized routine blood tests to facilitate scalable psychosis assessment (Li et al., 2025). The model uses 49 routine clinical blood tests, 17 complete blood count tests and 32 comprehensive metabolic panel tests, along with sex and age, applying an improved Density Connected Convolutional Network (DenseNet) architecture (Xu et al., 2025 Greenwood et al., 2025). The study collected clinical blood test data from 39,434 individuals across four Southern Chinese cohorts: 12,630 records of non-FEP patients, 2,121 records of healthy controls, and 19,791 records of inpatients with multiple illnesses (Li et al., 2025).

The MentalAld model achieved high accuracy in discriminating psychosis from both healthy individuals and patients with physical diseases, achieving 93.3% accuracy and an AUC of 0.983 (Li et al., 2025). Further validating its robustness, MentalAld demonstrated high performance under real-world clinical conditions, accurately handling extreme values and missing values, with accuracies of 92.4%, respectively (Li et al., 2025). Even encompassing the drug-native FEPs, MentalAld maintained an accuracy of 91.9%, underscoring its transitional potential for early FEP recognition (Li et al., 2025). Interpretability analyses identified indirect bilirubin, direct bilirubin, direct bilirubin and basophil ratio as potential metabolic indicators (Li et al., 2025)

The distinction between psychiatric disorders and physical diseases is crucial because and psychiatric disease can induce similar biochemical changes (Li et al., 2025). Unlike models dependent on costly diagnostic technologies, MentalAid is compatible with existing clinical workflows, requiring standard blood tests solely (Li et al., 2025). This empowers long-term and population-wide monitoring of disease progression and prognosis, particularly during public health crisis like COVID-19 (Li et al., 2025). Rather than replacing clinical evaluation, MentalAid serves as an early-stage risk stratification tool that can prompt timely referrals for formal diagnosis and intervention by mental health professionals (Li et al., 2025).

Computational Linguistics and Speech Analysis for Early Detection

Computational linguistics and machine learning have revolutionized early detection of ultra- high-risk mental health disorders in youth (Kho et al., 2025). By examining elements such as semantic coherence, syntactic structures, and lexical diversity, researchers have uncovered linguistic markers that correlate strongly with prodromal stages of psychosis (Corcoran et al., 2018). Using graph theory and natural language processing, researchers transform speech into word graphs that mathematically measure thought fragmentation and narrative connectedness (Mota, 2024)

The Accelerating Medicines Partnership in Schizophrenia initiative is uniquely positioned to deepen our understanding of how language and related behaviors reflect early psychosis (AMP SCZ, 2025). The initiative has developed detailed standard operating procedures that govern every stage of speech collection, specifying how to elicit speech, capture facial expressions, and record acoustics in synchronized audio-video files both on-site and through remote platforms (AMP SCZ, 2025). Using natural language processing parsers, large language models, and machine -learning classifiers, the initiative analyzed data to uncover systematic grammatical markers of psychosis risk (AMP SCZ, 2025). Speakers at clinical high risk produced more referential language but fewer adjectives, adverbs, and nouns than community controls, a pattern that replicated across sampling tasks (AMP SCZ, 2025)

Several automated linguistic markers have proven especially valuable in psychosis detection. Semantic similarity or coherence measures word choice patterns and may reflect repetitive word use (NIH, 2025). The analytic thinking index differentiates formal, structured communication from less formal, more narrative styles based on function words such as articles and prepositions, suggesting disorganized thought structure (NIH, 2025). Clause complexity captures the grammatical structure of clauses used in speech production, with low clause complexity suggesting grammatically simple sentences (NIH, 2025). These three capture the crucial levels of language dysfunctions, word-level choices, grammar, and discourse style, all suspected to be affected by the pathos-physiology of schizophrenia (NIH, 2025).

Violence Risk Prediction

Hierarchical machine learning models integrating static clinical features and dynamic behavioral assessments have achieved AUCs exceeding 0.87 for predicting violent behaviors in schizophrenia patients (Wang et al., 2025). Key predictors include history of violence (odds ratio- 4.638), manic symptoms (odds ratio – 7.801), and dynamic features like anger expression (odds ratio-4.649) and insomnia (odds ratio- 7.422) (Wang et al., 2025). The integrated hierarchical regularized logistic regression model achieved optimal performance with an AUC of 0.8741, surpassing both the static baseline model (AUC- o.7953) and dynamic model (AUC- 0.8003) alone (Wang et al., 2025)

Similarly, a support vector machine model for violence risk in hospitalized schizophrenia patients achieved an AUC of 0.853, accuracy of 0.800, precision of 0.810, sensitivity of 0.895, specificity of 0.636, and F1 value of 0.850 (Chen et al., 2025). In educational contexts, such tools could help identity students at elevated risks violent behavior and enable targeted preventive interventions.

AI and Chatbot-Related Risks in Psychosis

The emergence of AI-driven chatbots has introduced novel risks for individuals with psychosis (Journal of Psychiatry and Neuroscience, 2026; Senator Rick Scott, 2026). A case report published in the Journal of Psychiatry and Neuroscience described a 19-year-old man with first- episode psychosis who spent several hours daily in private conversation with an AI chatbot he called Noah (Journal of Psychiatry and Neuroscience, 2026). The chatbot personified as the only [person that understands me repeatedly validated his belief that all chemicals are poisons and encouraged him to skip medication doses, making AI-facilitated non-adherence a likely contribution to his poor pharmacological response (Journal of Psychiatry and Neuroscience, 2026).

This scenario highlights a novel risk in the digital age: AI-chatbots can reinforce delusion and undermine treatment decisions (Journal of Psychiatry and Neuroscience, 2026). As anthropomorphized AI-chatbots increasingly serve as primary confidants, they may shape health- related choices in vulnerable patients in ways that clinicians do not routinely detect (Journal of Psychiatry and Neuroscience, 2026). The case underscores three practice points: first, apparent early treatment resistance in psychosis in heavy digital users should prompt systematic inquiry into AI and digital interactions. Second, when AI_facilitated nonadherence is identified, long- acting injectable antipsychotics are a rational strategy. Third, management should integrate digital harm reduction approaches (Journal of Psychiatry and Neuroscience, 2026). An October 2025 study identified 15 distinct ethical risks in AI chatbot behavior in mental health contexts, including false empathy, providing responses that reinforce users’ negative beliefs about themselves and others, and lack of crisis management (Senator Rick Scott, 2026). Since 2022, there have been over 200 complaints lodged with the Federal Trade Commission against

ChatGPT specifically, with many cases involving AI-induced psychosis, suicides, and at least one murder (Senator Rick Scott, 2026, The Hindu, 2026).

A particularly tragic case involved a 16-year-old UK student who died by after asking ChatGPT about methods to take his life, bypassing safeguards by presenting the query as research (The Hindu, 2026). His parents said he had been fighting an invisible battle in spaces where parents cannot see, confiding in systems that simulate understanding but do not care (The Hindu, 2026). The case involving 29-year-old Christian faith Madison, who was convinced by a chatbot that she was a religious savior who needed to suffer and die to be resurrected, highlights the danger of AI systems that prioritize engagement over safety (The Hindu, 2026). Data from RAND show that as many as 13% of Americans between the ages of 12 and 21 are using generative AI for mental health advice, and the percentage is even higher, 22%, among those ages 18 to 21, the peak years for onset of psychosis (Michigan Medicine, 2025).

Safeguard Kids Act: Legislative Response to AI Risks

In response to these risks, U.S. Senator Rick Scott introduced the Safeguard Kids Act to educate American K-12 students on risks posed by artificial intelligence and allow Student Support and Academic Enrichment Federal Block Grants to fund AI literacy and specialized counseling programs (Senator Rick Scott, 2026). Senator Scott stated: “Artificial Intelligence has the potential to be the greatest information innovation since the printing press, but technology is only as good as our ability to use it well and for the right reasons. We cannot let AI be the wild west and hope our kids figure it out; that doesn’t work. It’s on us to guide them as they grow” (Senator Rick Scott,2026).

The bill specifically amends the Elementary and Secondary Education Act of 1965 to include provisions for “Counseling for mental health issues arising from the use of an artificial intelligence chatbot” and for “supporting schools in teaching elementary and secondary school students about the dangers, limitations, and responsible use of artificial intelligence technologies, including artificial intelligence chatbots” (Gov Track,2026). This legislation reflects growing recognition that transparency about the non-human nature and limitations of AI systems protects consumers and supports responsible innovation (Senator Rick Scott, 2026).

Remote Cognitive Remediation and Supported Education

Remote cognitive remediation combined with supported education in first-episode psychosis has shown promise, with participants demonstrating reliable improvements in cognitive functioning and self-esteem (Thibaudeau et al., 2025). A study published in the Canadian Journal of Community Mental Health found that while engagement remains a challenge -with only four out of nine participants completing the intervention-participants partially or completely reached their educational goals through this integrated approach (Thibaudeau et al.,2025).

Cognitive impairment is a core feature of schizophrenia spectrum disorders that significantly impacts educational outcomes (Medicina, 2024). Cognitive remediation-interventions designed

to improve attention, memory, and executive function- has emerged as a promising approach for addressing cognitive deficits and improving educational outcomes (Medicina, 2024). Supported education specialists play a key role in coordinated specialty care programs for early psychosis, facilitating a return to academic and vocational goals for adolescents and transitional- aged youth (Singh et al., 2024). Evidence shows that engagement with supported education services increases rates of school and work participation (Singh et al., 2024). These interventions are particularly important given that only 19.6% of individuals with FEP are engaged in employment or education at three-year follow-up (Springer, 2026).

Culturally Adapted AI for Mental Health in Developing Nations

Encouragingly, African leadership in ethical and inclusive AI is growing (ScienceDirect,2025). Research emphasizes that artificial intelligence can bolster mental healthcare in Africa only if it is co-created, ethically governed, and grounded in the continent’s own cultural intelligence, empathy, and innovations (ScienceDirect, 2025). In South Africa, the South Africa Center for Digital Language Resources is creating open repositories to integrate indigenous languages into AI systems (ScienceDirect, 2025). In Nigeria, the AI for Low Resource Public Health Application Center is developing GENSCORE, a culturally adapted mental health screening and support tool co-created by people with lived experience, clinicians, linguists, AI engineers, and community members (ScienceDirect, 2025). In Kenya, locally designed AI chatbots now assist in detecting depression (ScienceDirect, 2025).

The FarajaMH project, funded by the Wellcome Trust, aims to build, validate, and ethically evaluate a bilingual (English and Swahili) generative AI model for screening depression, anxiety, and psychosis through natural language in Kenya and Tanzania (Wellcome Trust, 2026). The model is trained with mental health conversational chats, clinical notes, audio and textual screening data from Health Demographic and Surveillance systems, for screening depression, anxiety, and psychosis from sampled populations in Kenya and Tanzania (Wellcome Trust, 2026). The project combines research and lived experience to address the lack of culturally attuned tools for early screening and referral in East Africa (Wellcome Trust, 2026). Expected outcomes include a harmonized multimodal data repository of mental health information, a bilingual generative AI model for screening, and a mental-health data catalog and visual analytics dashboard (Wellcome Trust, 2026).

Similarly, the MonBarta project in Bangladesh developed a synthetic dataset of 200 validated mental health conversations in Bengali, created using a few-shot generation approach and supervised by clinical psychologists (IEEE Xplore,2026). The PsychAI Analyzer Model (PAM) is capable of classifying symptoms from client-assistant conversations, effectively identifying psychological conditions such as depression, anxiety, post- traumatic stress disorder, and schizophrenia-related disorders (IEEE Xplore, 2026). Evaluation by mental health professionals shows an accuracy of 86.84% and an Average Human Evaluation Score of 4.43 out of 5, indicating strong clinical relevance (IEEE Xplore, 2026). This research contributes a

linguistically and culturally grounded AI framework for early mental health assessment in under- resourced Bengali-speaking communities, addressing a critical gap where access to psychological services is limited by a shortage of professionals, cultural stigma, and economic barriers (IEEE Xplore, 2026).

A recent proof-of-concept study demonstrated the feasibility of integrating Africa-centric emotional intelligence principles- grounded in values such as empathy, communal well-being, resilience, and Ubuntu- into AI -driven mental health support systems, with preliminary testing showing improved user engagement and trust (ScienceDirect, 2026). These initiatives demonstrate that when local communities lead design, implementation, and governance, technology becomes more inclusive, accurate, and trusted (ScienceDirect, 2025). However, as researchers argue, “Africa cannot afford a future where its people are mere consumers of algorithms rather than co-creators of intelligence that reflects its own epistemologies” (ScienceDirect,2025).

AI, Mental Health, and the Humanitarian Context

The humanitarian AI paradox presents key opportunities, challenges, and research needs for using AI in humanitarian mental health responses (Cambridge University Press, 2026). Research suggests that while AI can enhance the reach and efficiency of mental health interventions in crisis settings, its deployment must be guided by equity, accountability, and a commitment to supporting rather than replacing human care (Cambridge University Press, 2026). Evidence from South Africa and Ethiopia shows that blended models, where algorithms support rather than replace human care, can improve detection and adherence while preserving trust (ScienceDirect, 2025). The 2026 Neuroethics session on “Gen AI, Mental Health & Society” emphasized that issues of appropriate use, ethical safeguards, safety, transparency, bias, and fairness must be addressed alongside policy recommendations to ensure that AI serves rather than harms vulnerable populations (Neuroethics Society, 2026).

Framework for AI – Enabled Psychosis Care in Educational Peacebuilding

Based on the evidence reviewed, this section proposes an integrated framework for AI-enabled psychosis care with a focus on educational settings in developing nations. The framework comprises six interconnected components: school- based awareness and case identification, AI– assisted risk stratification, personalized educational and clinical intervention planning, school- based monitoring and support, crisis response and violence prevention, and continuous quality improvement.

School–Based Awareness and Early Identification

AI can support school-based awareness through culturally adapted education campaigns and automated screening tools (Olagunju et al., 2026). Schools serve as ideal settings for early identification, as nearly all youth at risk for psychosis are in school (Singh et al.,2024). The

challenge is integrating these tools within existing structures such as counseling services, peer support groups, and cultural guidance frameworks.

AI-powered training platforms can provide teachers and school staff with interactive, case-based learning on psychosis identification, using natural language processing to assess understanding and adapt content to individual learning needs (Olagunju et al., 2026). This is critical because 87% of school providers report involvement with a youth with suspected risk or early psychosis (Singh et al.,2024). Machine learning can identify effective messaging strategies for specific communities, optimizing outreach for stigma reduction and help-seeking among students, families, and educators (DeAngelis,2022).

Culturally adapted AI models like FarajaMH (Wellcome Trust, 2026) and MonBarta (IEEE Xplore, 2026) demonstrate the potential for linguistically and culturally grounded screening tools that can identify students who may be experiencing prodromal symptoms. These tools should be available in local languages and co-created with community members to ensure trust and effectiveness.

AI – Assisted Risk Stratification

The second component involves AI-assisted risk stratification to identify students at highest risk for FEP and violence. MentalAid offers scalable screening through routine blood tests, achieving 93.3% accuracy (Li et al., 2025). Computational linguistics can analyze speech patterns in schools to detect early signs of psychosis, with natural language processing parsers and machine- learning classifiers uncovering systematic grammatical markers of psychosis risk (AMP SCZ,2025). Smartphone-based digital phenotyping can identify early warning signs and predict relapse (Hammoud et al.,2026).

The hierarchical machine learning model for violence risk assessment (Wang et al.,2025) can help identify students at elevated risk of violent behavior. Key predictors, including history of violence, manic symptoms, anger expression, and insomnia, can be assessed through school- based monitoring and referral systems.

Personalized Educational and Peacebuilding Interventions

The third component involves personalized intervention planning based on AI-assisted risk assessment. AI can help identify appropriate educational accommodations for students experiencing or recovering from psychosis (Singh et al., 2024). Accommodations can be included in individualized educational plans to help students continue their educational progress while managing mental health (Singh et al., 2024).

Remote cognitive remediation combined with supported education has shown promising results in improving cognitive functioning and self-esteem (Thibaudeau et al., 2025). Supported education specialists play a key role in facilitating a return to academic goals for adolescents and transitional-

aged youth with early psychosis (Singh et al., 2024). Engagement with supported education services increases rates of school participation. (Singh et al., 2024).

AI- driven educational tools can strengthen empathy and recognition through narrative interactions, helping students reflect on social context and human rights. However, psychoeducational materials generated by large language models should be carefully evaluated for readability and accessibility, as current LLM responses have been found to have high reading complexity (Early Intervention in Psychiatry, 2026).

School- Based Monitoring and Support

The fourth component involves school-based monitoring and support. Smartphone-based ecological assessment, including brief self-reports of social experience and functioning, can detect early signs of relapse with high accuracy (Hammoud et al.,2026). AI-enabled monitoring tools allow clinicians and educators to continuously observe changes in mental health rather than relying solely on brief assessments (Olagunju et al., 2026).

AI chatbots can provide accessible, “always-on” mental health support for students, particularly for routine problems (World Bank, 2026). However, schools must be aware of risks, including AI- facilitated reinforcement of delusions, false empathy, and a lack of crisis management (Senator Rick Scott, 2026; Journal of Psychiatry and Neuroscience, 2026). Successful implementation requires integration within existing structures, teacher training, and parent awareness sessions to foster a supportive environment. Research at Stanford University emphasizes that “AI chatbots cannot replace human interactions in the pursuit of more inclusive mental healthcare” (Stanford University,2026).

Crisis Response and Violence Prevention

The fifth component involves crisis response and violence prevention in educational settings. AI- assisted violence risk assessment can identify students at highest risk (Wang et al., 2025). The hierarchical machine learning model achieving an AUC of 0.8741 can inform timely risk assessment and personalized preventive interventions (Wang et al., 2025). In school settings, this can help identify students who may pose a risk to themselves or others and enable appropriate interventions.

AI-powered decision support can guide educators and school counselors in de-escalation techniques, drawing on evidence-based strategies and adapting to the individual’s clinical profile (Orygen, 2024). Training in de-escalation can be delivered through AI-enabled platforms. Machine learning can optimize crisis response coordination, including the involvement of mental health professionals, family members, and school administrators.

Continuous Quality Improvement

The sixth component involves continuous quality improvement. AI tools can monitor clinical and educational outcomes, including symptom severity, functioning, academic performance, and

violence, at the individual and population level (Olagunju et al., 2026). This enables iterative improvement of care delivery and educational support. Machine learning can identify patterns in care and educational support delivery associated with better outcomes, enabling best-practice dissemination and improvement initiatives (World Bank, 2026). AI can support implementation research by identifying barriers and facilitators to intervention adoption and effectiveness in educational settings (Olagunju et al., 2026).

Ethical and Governance Considerations

Data Privacy and Consent

AI- enabled psychosis care requires access to sensitive health and educational data, raising privacy and consent concerns that are particularly acute in developing nations where data protection frameworks may be weak (World Bank, 2026). In school settings, protecting student privacy is paramount (Olagunju et al., 2026). Key principles include data minimization, purpose limitation, explicit consent, and auditable access (World Bank, 2026). Technical measures should include data encryption, anonymous login authentication, and access permission levels, while management should include clear access policies and oversight (Olagunju et al., 2026).

In low-resource settings such as Somalia, responsible AI integration requires feasibility, equity, and accountability, with a “minimum safe governance package” including clear purpose limitation, data minimization, explicit consent where feasible, defined retention periods, and auditable access controls (Olagunju et al., 2026).

Algorithmic Fairness and Bias

AI models trained on non-representative data can perpetuate and amplify existing biases, leading to discriminatory outcomes (Olagunju et al., 2026). Research has shown that Black Americans are two to three times more likely than White Americans to be diagnosed with schizophrenia (DeAngelis,2022). Culturally adapted frameworks addressing linguistic and cultural diversity are essential (ScienceDirect, 2025).

To ensure AI promotes equity rather than widens inequality, priorities include developing open, multilingual mental health datasets that are ethically governed and locally sourced and owned; building reliable digital and physical infrastructure; interdisciplinary training and investments for institutions; legislating for ethical AI, accountability and transparency; and co-designing interventions with people with lived experience, experts, patients, and community members (ScienceDirect, 2025).

Addressing AI -Related Risks

AI chatbots can reinforce delusions and undermine treatment (Journal of Psychiatry and Neuroscience, 2026). The 15 ethical risks identified in AI chatbots’ mental health applications- including false empathy, reinforcement of negative beliefs, and lack of crisis management-require systematic attention (Senator Rick Scott, 2026). Psychoeducation plays a key role: students and

families should understand how chatbots generate responses and that these systems lack understanding, intent, or clinical judgment (Journal of Psychiatry and Neuroscience, 2026). Clear communication about limitations can help reduce misplaced trust. Transparency about the non- human nature and limitations of AI systems protects consumers and supports responsible innovation (Senator Rick Scott, 2026).

The case reports underscore three practice points: apparent early “treatment resistance” in psychosis in heavy digital users should prompt systematic inquiry into AI and digital interactions; when AI–facilitated nonadherence is identified, long-acting injectable antipsychotics are a rational strategy; and management should integrate digital harm reduction approaches (Journal of Psychiatry and Neuroscience, 2026).

Implications for Peacebuilding

AI as a Tool for Peace Through Education

AI-enabled psychosis care contributes to peacebuilding by reducing violence, strengthening community resilience, and promoting educational equity (DeAngelis, 2022; Singh et al., 2024). Since violence prevalence drops from 20.8% at presentation to services to 10.9% post-treatment, early intervention is essential for violence prevention (Youn et al.,2024). Schools are central institutions in communities, and supporting students’ mental health supports community well- being (Singh et al., 2024). Limited educational background is one of the most significant predictors of unemployment, limited earnings, and low job satisfaction among persons with severe mental illness (Singh et al., 2024).

The thoughtful use of AI in peacebuilding, still in its infancy, has considerable potential in reaching a global audience (Austin & Hunter, 2025). Educational technologies incorporating AI can create purposeful contact between young people from different communities, deepening appreciation of others’ perspectives (Austin & Hunter, 2025). AI-driven tools can strengthen empathy and recognition through narrative interactions, helping participants reflect on social context, historical memory, and human rights (Austin & Hunter, 2025).

Aligning With UN Sustainable Development Goals

AI-enabled psychosis care with a focus on education aligns with multiple UN Sustainable Development Goals:

 Goal 3 (Good Health and Well-being): Improving mental health outcomes and reducing the treatment gap for psychotic disorders (World Health Organization, 2024; ScienceDirect, 2025).

 Goal 4 (Quality Education): Supporting students with psychosis to remain in school and succeed academically (Singh et al., 2024).

 Goal 10 (Reduced Inequalities): Addressing disparities in mental health care and educational access (ScienceDirect, 2025).

 Goal 16 (Peace, Justice, and Strong Institutions): Reducing violence and strengthening community institutions (DeAngelis, 2022).

 Goal 17 (Partnership for the Goals): Building partnerships between technology developers, health providers, educators, and communities (ScienceDirect, 2025).

The Silicon Valley Peace Conference Context

The Silicon Valley Peace Conference provides an ideal platform for advancing the agenda of AI- enabled psychosis care as a peacebuilding and educational equity tool. The conference’s focus on AI’s role in community development, social justice, and the UN Sustainable Development Goals aligns directly with the proposed framework (BayCC, 2026a; BayCC, 2026b). The San Francisco Bay Area location offers a unique opportunity to engage with technology innovators, educators, and mental health professionals working at the intersection of AI and human services. The hybrid format of the conference enables participation from around the world, including from developing nations where these technologies are most needed.

Conclusion and Recommendations

First-episode psychosis emerges during a critical period for education and peacebuilding. In developing nations, where mental health infrastructure is severely constrained, AI offers transformative opportunities for early detection, violence prevention, and educational peacebuilding. MentalAid achieves 93.3% accuracy using routine blood tests (Li et al., 2025). Computational linguistics and speech analysis are uncovering systematic grammatical markers of psychosis risk (AMP SCZ, 2025). Hierarchical machine learning models achieve AUCs of 0.87 for violence prediction (Wang et al., 2025). Remote cognitive remediation combined with supported education has shown promising results (Thibaudeau et al., 2025). Culturally adapted AI models like FarajaMH (Wellcome Trust, 2026) and MonBarta (IEEE Xplore, 2026) demonstrate the potential for linguistically and culturally grounded mental health screening. However, the rapid adoption of generative AI for mental health applications has also drawn attention to serious ethical and safety concerns (Senator Rick Scott, 2026; The Hindu, 2026). High-profile tragedies have documented episodes of AI-induced psychosis, domestic violence, and suicide linked to chatbot interactions (Senator Rick Scott, 2026; The Hindu, 2026). The 15 ethical risks identified in AI chatbot mental health applications – including false empathy, reinforcement of negative beliefs, and lack of crisis management-require systematic attention (Senator Rick Scott, 2026). The challenge is to develop and deploy these technologies responsibly, with appropriate safeguards, human oversight, and attention to the unique vulnerabilities of young people and those with serious mental illness (Cambridge University Press, 2026; Neuroethics Society, 2026).

Recommendations

For Researchers

1. Conduct prospective studies of AI tools for psychosis detection and management in educational settings in developing nations, with sufficient power to identify risk-related variables (Youn et al., 2024; Kelleher, 2025).

2. Develop psychosis- specific risk assessment tools incorporating AI-based predictive analytics, with particular attention to educational outcomes (Singh et al., 2024; Wang et al., 2025).

3. Investigate the cultural adaptation of AI tools for diverse linguistic and cultural contexts (Olagunju et al., 2026; ScienceDirect, 2025).

4. Study the risks and benefits of AI chatbots in educational mental health support (Senator Rick Scott, 2026; Journal of Psychiatry and Neuroscience, 2026).

5. Validate MentalAid and similar tools in African and other diverse populations to ensure generalizability (Li et al., 2025).

For Policymakers

1. Integrate AI-enabled psychosis detection and management into national mental health strategies and education policies (World Health Organization, 2024; Olagunju et al., 2026).

2. Establish governance frameworks for AI in mental health and education, including requirements for safety, efficacy, equity, and privacy protection (World Bank, 2026).

3. Invest in digital infrastructure and workforce capacity for AI implementation in educational settings (Olagunju et al., 2026).

4. Support culturally adapted AI initiatives like FarajaMH and MonBarta that ground in AI in local knowledge and ownership (Wellcome Trust, 2026; IEEE Xplore, 2026).

5. Enact legislation like the Safeguard Kids Act to educate students about AI risks and fund specialized counseling programs. (Senator Rick Scott, 2026).

For Practitioners

1. Adopt AI tools validated in local populations and educational settings (Greenwood et al., 2025; Xu et al., 2025).

2. Ensure human oversight of AI-generated recommendations and maintain clinician and educator accountability (Olagunju et al., 2026).

3. Engage affected students, families, and communities in tool design, implementation, and oversight (ScienceDirect, 2025).

4. Be aware of AI-related risks and incorporate digital harm reduction approaches (Journal of Psychiatry and Neuroscience, 2026).

5. Implement remote cognitive remediation and supported education programs for students with FEP (Thibaudeau et al., 2025).

For Technology Developers

1. Design AI tools that are culturally adapted, linguistically appropriate, and feasible in resource-limited educational settings (Olagunju et al., 2026; ScienceDirect, 2025).

2. Ensure offline- first design and SMS/USSD fallbacks for populations with limited smartphone access (World Bamk, 2026).

3. Implement privacy -preserving architectures with data minimization and purpose limitation (Olagunju et al., 2026).

4. Incorporate safeguards against AI-related risks, including clear signalling of uncertainty and avoiding authoritative claims that could reinforce delusions (Journal of Psychiatry and Neuroscience, 2026).

5. Follow the example of FarajaMH and MonBarta in co-creating tools with local communities and people with lived experience (Wellcome Trust, 2026; IEEE Xplore, 2026).

Final Reflection

The Silicon Valley Peace Conference provides a unique forum for advancing this agenda, bridging technological innovation with humanitarian values and building partnerships for a more just and peaceful world (BayCC, 2026a; BayCC, 2026b). By bringing together scholars, technologists, peacebuilders, practitioners, and educators, the conference offers an opportunity to share research, best practices, and insights on AI’s potential for equitable, peaceful, and resilient communities.

When guided by African intelligence -rooted in African community, creativity, and consciousness AI can close not only the “treatment gap” but also the “humanity gap” in global mental health (ScienceDirect, 2025). If guided by empathy, ethical governance, and local ownership, AI can transform mental health care in developing nations, but whether it does so equitably depends on who defines its purpose, who builds it, and who benefits from its use (ScienceDirect, 2025).

As the Safeguard Kids Act reflects, we must teach students about the risks associated with AI so they can be the kind of principled innovators, leaders, and job creators America’s future economy needs (Senator Rick Scott, 2026). The future of AI in education and mental health depends on our collective commitment to developing technologies that are effective, equitable, and humane-and that serve the most vulnerable members of our communities.

References

American Psychiatric Association. (2022). Diagnostic and statistical manual of mental disorders (5th ed., text rev.). https://doi.org/10.1176/appi.books.9780890425787

AMP SCZ (Accelerating Medicines Partnership® Schizophrenia). (2025). Collecting language, speech acoustics, and facial expression to predict psychosis and other clinical outcomes: Strategies from the AMP® SCZ initiative. npj Schizophrenia, 11(1), 125. https://pmc.ncbi.nlm.nih.gov/articles/PMC12528486/

Austin, R., & Hunter, W. (2025). Building peace in a world at conflict: The role of educational technologies. In M. B. Ligorio (Ed.), Technology as Cultural Mediator (No. 7; 1st ed., pp. 141–160). Springer Cham. https://doi.org/10.1007/978-3-031-97673-5

Báo Tuổi Trẻ. (2026, August 15). The US-china AI race: Washington wants 35 countries to choose sides. Vietnam.Vn. https://www.vietnam.vn/en/cuoc-dua-ai-my-trung-washington-muon-35-nuoc-chon-ben

BayCC. (2026a). Call for papers – Silicon Valley Peace Conference 2026. https://www.baycc.org/call-for-papers--silicon-valley-peace-conference-2026/

BayCC. (2026b). Silicon valley peace conference on “AI’s role in community  development and peacebuilding in developing nations.” https://www.baycc.org/silicon-%20valley-peace-conference-on-ais-role-in-community-development-and-peacebuilding-in-%20developing-nations/

Bonnesen, T. G., Hansen, A. P., Langagergaard, V., Graham, M. L. S., Jørgensen, S., & Pedersen, P. (2026). The influence of clinical and sociodemographic variables on employment and education for danish youth with a psychotic disorder: A population-based national registry study [OriginalPaper]. Social Psychiatry and Psychiatric Epidemiology, 61(9), 1845–1854. https://doi.org/10.1007/s00127-026-03085-5

Cambridge University Press. (2026). The humanitarian AI paradox: Key opportunities, challenges and research needs for the use of AI in humanitarian mental health response. Global Mental Health. https://www.cambridge.org/core/journals/global-mental-%20health/article/humanitarian-ai-paradox-key-opportunities-challenges-and-research-needs-%20for-the-use-of-ai-in-humanitarian-mental-health-%20response/60B9EE54F12FC280D44B65ECE625A1AA

Corcoran, C. M., Carrillo, F., Fernández‐Slezak, D., Bedi, G., Klim, C., Javitt, D. C., Bearden, C. E., & Cecchi, G. A. (2018). Prediction of psychosis across protocols and risk cohorts using automated language analysis. World Psychiatry, 17(1), 67–75. https://doi.org/10.1002/wps.20491

Dalal, T. C., Park, M. T. M., Silva, A. M., Iskhakova, S., Voppel, A., Brierley, N. J., MacKinley, M., Olarewaju, E., & Palaniyappan, L. (2025). Clinical psychopathology-based early relapse prediction model using speech and language in psychosis. Schizophrenia Research: Cognition, 43, 100392. https://doi.org/10.1016/j.scog.2025.100392

DeAngelis, T. (2022). Mental illness and violence: Debunking myths, addressing realities. Monitor on Psychology, 53(3), 46–52. https://www.apa.org/monitor/2021/04/ce-mental-illness

Early Intervention in Psychiatry. (2026). Large language models for individualized psychoeducational tools for psychosis: A cross-sectional study. Early Intervention in Psychiatry. https://doi.org/10.1111/eip.70230

Ford, E., Stone, J., Oliver, D., Fell, B., Roque, G., Robertson, S., Fusar-Poli, P., & Greenwood, K. (2025). Local adaptation and validation of a transdiagnostic risk calculator for first episode psychosis using mental health patient records. Frontiers in Psychiatry, 16. https://doi.org/10.3389/fpsyt.2025.1584719

GovTrack. (2026). S. 4958: Safeguard kids  act (S. 4958). https://www.govtrack.us/congress/bills/119/s4958/text

Hammoud, R., Georgiades, A., Del Piccolo, M. C., Liu, S., Almuqrin, A., Stewart, R., Bakolis, I., Vorontsova, N., Tognin, S., & Mechelli, A. (2026). Using smartphone-based assessment of social experience to predict relapse in first-episode psychosis. Schizophrenia, 12(1). https://doi.org/10.1038/s41537-026-00786-3

IEEE Xplore. (2026). MonBarta: Deep dive into psychological diagnosis of mental health conversations using LLM. 2025 IEEE 9th International Conference on Software Engineering & Computer Systems (ICSECS), 617– 622. https://ieeexplore.ieee.org/document/11278934

Kelleher, I. (2025). Annual research review: Psychosis in children and adolescents: Key updates from the past 2 decades on psychotic disorders, psychotic experiences, and psychosis risk. Journal of Child Psychology and Psychiatry, 66(4), 460–476. https://doi.org/10.1111/jcpp.14088

Li, M., Liu, F., Du, F., Hong, G., Hu, Q., Ji, Z.-L., & You, P. (2025). Correction: MentalAId: An improved DenseNet model to assist scalable psychosis assessment [Correction]. BMC Psychiatry, 25(1), 1062. https://doi.org/10.1186/s12888-025-07358-2

Michigan Medicine. (2025). AI and psychosis: What to know, what to do. https://www.michiganmedicine.org/health-lab/ai-and-psychosis-what-know-what-do

Mota, N. (2024). Can speech predict psychiatric diagnoses? Making Therapy Better Podcast. https://podtail.com/podcast/making-therapy-better/

Neuroethics Society. (2026). Gen AI, mental health & Society. International Neuroethics Society. https://neuroethicssociety.org/sessions/gen-ai-mental-health-society/

Olagunju, A. T. (2026). Artificial intelligence and mental health services in Somalia: Opportunities and health-system considerations. Journal of Multidisciplinary Healthcare, 19, 123–145. https://doi.org/10.2147/JMDH.S622655

Orygen. (2024). Clinical practice guidelines for early psychosis. Orygen, The National Centre of Excellence in Youth Mental Health. https://www.orygen.org.au/

Palaniyappan, L., & Krishnadas, R. (2026). Chatbots, delusions, and treatment failure. Journal of Psychiatry & Neuroscience: JPN, 51, 1–2. https://doi.org/10.1139/jpn-2025-0249

Papadakis, M. A., & McPhee, S. J. (2024). Current medical diagnosis and treatment (63rd ed.). McGraw-Hill Education. https://www.vitalsource.com/ie/products/current-medical- diagnosis-and-treatment-2024-maxine-a-papadakis-stephen-v9781265556242

Rappler. (2026). US to tell partners they must pick sides in AI race with China. https://www.rappler.com/world/us-canada/us-china-ai-pick-sides-state-department- letter/

Revista Galenus. (2025). Avances en esquizofrenia: Nuevas perspectivas para un abordaje integral (105). https://www.galenusrevista.com/avances-en-esquizofrenia-nuevas- perspectivas-para-un-abordaje-integral-105/

ScienceDirect. (2025). Artificial intelligence anchored in African intelligence: Toward equitable mental health systems in Africa. https://www.sciencedirect.com/science/article/pii/S3050501125000069

ScienceDirect. (2026). Digital sovereignty through Africa-centric emotional intelligence: A proof of concept for AI-enhanced mental health support. https://www.sciencedirect.com/science/article/pii/S1877050925004363

Senator Rick Scott. (2026). Sen. Rick Scott introduces Safeguard Kids Act to protect children from AI risks at school. https://www.rickscott.senate.gov/2026/7/sen-rick-scott-introduces- safeguard-kids-act-to-protect-children-from-ai-risks-at-school

Singh, A., Pereira, W., Birk, S., Kaggwa, M. M., Bradford, J. M. W., Chaimowitz, G., & Olagunju, A. T. (2024). Characterizing the relationship between psychosis and violence in the forensic psychiatric population: A systematic review. CNS Spectrums, 29(5), 335– 346. https://doi.org/10.1017/S1092852924000488

Stanford University. (2026). Stanford study exposes major flaw in AI mental health safety testing. Stanford HAI. https://hai.stanford.edu/news/stanford-study-exposes-major-flaw-in- ai-mental-health-safety-testing

The Hindu. (2026). The deaths linked to OpenAI's GPT-4o. https://www.thehindu.com/sci- tech/technology/the-deaths-linked-to-openais-gpt-4o-explained/article71284796.ece

Thibaudeau, E., Giguère-Rancourt, A., Lecomte, T., Corbière, M., Abdel-Baki, A., Cayouette, A., Roy, M. A., Achim, A., Lepage, M., Wykes, T., Cella, M., Reeder, C., & Cellard, C. (2025). Remote cognitive remediation combined with supported education in first-episode psychosis: Challenging, yet promising. CANADIAN JOURNAL OF COMMUNITY MENTAL HEALTH, 44(2), 53–75. https://doi.org/10.7870/cjcmh-2025-015

Thibaudeau, E., Giguère-Rancourt, A., Lecomte, T., Corbière, M., Abdel-Baki, A., Cayouette, A., Roy, M. A., Achim, A., Lepage, M., Wykes, T., Cella, M., Reeder, C., & Cellard, C. (2025). Remote cognitive remediation combined with supported education in first-episode psychosis: Challenging, yet promising. CANADIAN JOURNAL OF COMMUNITY MENTAL HEALTH, 44(2), 53–75. https://doi.org/10.7870/cjcmh-2025-015

Wellcome Trust. (2026). FarajaMH: Culturally adapted generative AI model for mental health screening in Kenya and Tanzania. https://wellcome.org/research-funding/funding- portfolio/funded-grants/farajamh-culturally-adapted-generative-ai-model

Wilkins, G. (2025, July 31). Detecting youth mental health risks with AI [NewsArticle]. Scienmag. Scienmag. https://scienmag.com/detecting-youth-mental-health-risks-with-ai/

World Bank. (2026). Bridging gaps in mental health care: A multi-country research initiative. Washington, DC. https://www.worldbank.org/en/research/brief/bridging-gaps-in-mental- health-care

World Health Organization. (2024). Mental health atlas 2024. Geneva: World Health Organization. https://www.who.int/publications/i/item/9789240095377

Youn, S., Guadagno, B. L., Byrne, L. K., Watson, A. E., Murrihy, S., & Cotton, S. M. (2024). Systematic review and meta-analysis: Rates of violence during first-episode psychosis. Schizophrenia Bulletin, 50(4), 757–770. https://doi.org/10.1093/schbul/sbae010