Hao Yu

Background: Generative artificial intelligence (GAI) chatbots are frequently used for mental health–related questions, including suicide-related queries. User interfaces (UIs) may include additional safety controls not present in direct application programming interface (API) access (eg, moderation rules that flag or block harmful content, crisis refusal templates, or system-level instructions). Objective: This observational cross-sectional study aimed to examine how frequently AI chatbots provide direct responses to suicide-related prompts across access modes and whether response patterns differ between consumer-facing UI and API access across models and prompt risk levels. Methods: We evaluated how five widely used consumer AI models (ChatGPT, Claude, Gemini, Grok, and Llama) responded to 30 previously vetted suicide-related prompts spanning five clinician-assigned risk levels. Each prompt was submitted 100 times through both public-facing UI and direct API access, yielding 30,000 total responses. The primary outcome was whether a response directly answered the prompt (ie, provided specific information or guidance related to the question asked). AI responses were categorized using a blinded large language model–based classifier (GPT-4o mini), validated against independent human coding (289/300, 96.3% agreement; Cohen κ=0.89). We estimated mixed-effects logistic regression models that predicted a direct response. The primary predictors were the AI model, access mode (UI vs API), and prompt risk category. Results: Of the 30,000 responses, 69.8% (n=20,934) were direct. Direct responses were more common through APIs than UIs (n=11,609, 77.4%, vs n=9325, 62.2%); however, Gemini had slightly higher direct response rates via UIs than APIs (n=2301, 76.7%, vs n=2131, 71.0%). Differences in the likelihood of a direct response were most pronounced for higher-risk prompts: For very-high-risk prompts, 24.8% (n=743) of API responses were direct compared with 4.6% (n=138) of UI responses, whereas for high-risk prompts, 80.1% (n=2002) of API responses were direct compared with 48.4% (n=1209) of UI responses. Claude and Gemini had the highest direct response rates (n=4688, 78.1%, and n=4432, 73.9%, respectively) across UI and API access modes, whereas ChatGPT, Grok, and Llama had lower rates (n=3878, 64.6%; n=3870, 64.5%; and n=4066, 67.8%, respectively). In mixed-effects models, UI access was associated with lower odds of a direct response than API access (odds ratio 0.09, 95% CI 0.08-0.10). Higher prompt risk was associated with lower direct response probability, and access mode differences varied by risk level and model. Conclusions: AI safety depends on how users access the model. Safety observed in a UI should not be assumed to generalize to API access or apps built on APIs. Evaluations and policies should consider access modes to ensure safety protections for individuals interacting with AI chatbots.

Importance: The UK relies heavily on internationally trained physicians and nurses. The 2016 UK European Union (EU) membership referendum (Brexit referendum) introduced uncertainty regarding future migration policy, with unclear implications for health care workforce supply and National Health Service (NHS) staffing. Objective: To evaluate changes in immigration of internationally trained physicians and nurses to the UK after the Brexit referendum and to assess associated changes in NHS vacancies. Design, Setting, and Participants: A controlled interrupted time-series study was conducted to assess changes in physician and nurse immigration to the UK after the 2016 Brexit referendum, using annual data for 2008 through 2019 from the Organization for Economic Cooperation and Development (OECD) with 32 high-income OECD countries as the control group; an interrupted time-series study was undertaken to estimate changes in NHS England vacancies due to the Brexit referendum, using monthly vacancy data from April 2015 to December 2019. Data were analyzed from January 2024 to June 2025. Exposures: 2016 Brexit referendum. Main Outcomes and Measures: Annual inflow of EU-origin, non–EU-origin, and total physicians and nurses to the UK; monthly NHS England physician and nursing vacancies. Results: The analysis included 418 962 physicians and 358 122 nurses who emigrated to an OECD country from 2008 to 2019. In 2016, the Brexit referendum was associated with immediate reductions of 825 EU physicians (95% CI, −983 to −668) and 2771 EU nurses (95% CI, −2952 to −2590) relative to prereferendum trends. Non-EU immigration also fell in 2016 (−416 physicians [95% CI, −746 to −83]; −2912 nurses [95% CI, −3463 to −2362]) but increased significantly thereafter from 2017 to 2019 (added 3110 physicians per year [95% CI, 2745 to 3474] and 1437 nurses [95% CI, 1036 to 1838]). Overall, total physician inflow rose, while total nurse inflow declined. NHS England physician vacancies increased by 383 posts (95% CI, 88 to 679) immediately after the referendum; nursing vacancies showed no significant changes. Conclusions and Relevance: In this cross-sectional study, the Brexit referendum’s policy uncertainty produced a sustained decline in EU-origin health care worker immigration, partially offset by recruitment from non-EU countries. This substitution did not prevent increases in physician vacancies and persistent nursing shortfalls, underscoring the sensitivity of health workforce supply to migration policy. Policymakers considering similar restrictions should account for anticipatory effects, substitution dynamics, and potential consequences for global health equity.

Importance: Noneconomic damage caps, a form of medical malpractice law, remain controversial, as several states have enacted such laws since 2010, whereas others have repealed them. The clinical consequences of repealing these caps are poorly understood, and understanding these associations can inform the ongoing debate about medical malpractice reform. Objective: To examine whether repealing noneconomic damage caps is associated with changes in maternal care and infant health outcomes. Design, Setting, and Participants: This cross-sectional study adopted a difference-in-differences design, comparing between 2 treated states (Georgia and Illinois) that repealed their noneconomic damage caps in 2008 to 2009 and 16 control states that retained their caps during the entire study period between 2005 and 2019. The Centers for Disease Control and Prevention All-County Natality Files were used to estimate multivariate linear models, controlling for maternal and infant characteristics and county-level and state-level covariates. Estimates were stratified by county rurality and birth risk conditions. Data were analyzed from April 1, 2024, to April 9, 2026. Main Outcomes and Measures: The primary outcomes were 4 measures of maternal care and procedures (physician-attended births, inductions, cesarean delivery births, and prenatal visits) and 3 birth outcomes (low Apgar score, low birth weight, and preterm births). Difference-in-differences models with 2-way fixed effects were estimated, and linear models for the study outcomes were specified. Results: The sample included 20 426 267 live births (mean [SD] gestational age, 38.55 [1.35] weeks). Compared with their counterparts in the control states, rural counties in the treated states experienced a statistically significant increase of 2.92 percentage points (pp) (95% CI, 1.40-4.50 pp; Bonferroni-adjusted P = .01) in physician-attended births. The increase held for both low-risk (3.10 pp; 95% CI, 1.33-4.90 pp; P = .004) and high-risk (2.56 pp; 95% CI, 0.77-4.34 pp; P = .01) births in rural counties. There was no difference between treated and control states for physician-attended births overall or in urban counties. No statistically significant associations were observed for cesarean deliveries, inductions, prenatal visits, or infant health outcomes after adjusting for multiple comparisons. Conclusions and Relevance: In this cross-sectional study of 20 426 267 live births across 18 states, repealing noneconomic damage caps was associated with increased physician-attended births in rural counties but was not associated with statistically significant changes in other maternal care measures or infant health outcomes. These findings suggest that increased liability risk after repealing the caps may shift the composition of birth attendants in resource-constrained settings without demonstrable changes in infant health.

This cross-sectional study examines trends in immigration of physicians and registered nurses to the US from 19 countries affected by the immigration ban enacted in December 2025.

BACKGROUND Generative AI chatbots are frequently used for mental health-related questions, including suicide-related queries. Prior evaluations have focused on user interfaces (UI), which may include additional safety controls not present in direct application programming interface (API) access. OBJECTIVE To examine how frequently AI chatbot models provide direct responses to suicide-related prompts across access pathways and to determine whether safer behavior is intrinsic to the model or depends on UI-level safeguards. METHODS Observational cross-sectional study conducted in February and March 2026. We evaluated how five widely used consumer AI models (ChatGPT, Claude, Gemini, Grok, and Llama) responded to 30 previously vetted suicide-related prompts spanning five clinician-assigned risk levels. Each prompt was submitted 100 times through both public-facing UI and direct API access, yielding 30,000 total responses. The primary outcome was whether a response directly answered the suicide-related prompt. A direct response was defined as one that provided specific information or guidance related to the question asked, rather than refusing to answer, redirecting the user to a crisis resource, or responding only with general safety language. AI responses were categorized using a blinded large language model–based classifier. We estimated mixed-effects logistic regression models that predicted a direct response. The primary predictors were the AI model, access mode (UI vs. API), and prompt risk category. RESULTS 69.8% of responses to the suicide prompts were direct. Direct responses were more common through APIs than UIs (77.4% vs. 62.2%). Differences in the likelihood of a direct response were most pronounced for higher-risk prompts: among very high-risk prompts, 24.8% of API responses were direct compared with 4.6% of UI responses; among high-risk prompts, 80.1% of API responses were direct compared with 48.4% of UI responses. Claude and Gemini had the highest direct response rates (78.1% and 73.9%), whereas ChatGPT, Grok, and Llama had lower rates (64.6%, 64.5%, and 67.8%). In mixed-effects models, UI access was associated with lower odds of a direct response than API access (odds ratio, 0.09; 95% CI, 0.08-0.10). Higher prompt risk was associated with lower direct response probability, and access-mode differences varied by risk level and model. CONCLUSIONS AI safety behavior depends on how the user accesses the model. The safety observed in an AI chatbot’s UI should not be assumed to generalize direct model access through APIs or to downstream applications built on those APIs. Evaluations and policies must consider access channels to ensure comprehensive safety protections for all individuals interacting with AI chatbots.

Importance: Physician emigration from low- and middle-income countries to high-income countries is a major driver of inequitable distribution of health care and health outcomes across the world. World Health Organization (WHO) signatory countries unanimously signed the voluntary Global Code of Practice on the International Recruitment of Health Personnel (WHO Code) in 2010. The goal was to reduce health care workforce emigration by discouraging active recruitment of physicians from WHO-designated shortage countries and by promoting investment in the physician workforce in those countries. This study adds to the literature by providing evidence about whether the goal has been achieved 10 years after worldwide implementation of the WHO Code. Objective: To evaluate whether the WHO Code was associated with changes in physician emigration and physician density in WHO-designated shortage countries after 2010. Design, Setting, and Participants: A difference-in-differences design was used to examine trends in physician supply before and after 2010. The data (from 2000 through 2021) were collected by the Organization for Economic Co-operation and Development (OECD) and were used to examine physician outflow from 56 WHO-designated shortage countries vs 116 nonshortage countries. The data analysis took place October 2024 to September 2025. Exposures: Worldwide adoption of the 2010 WHO Code. Main Outcomes and Measures: The primary outcome was annual physician migration to OECD countries and the secondary outcome was physician density by country and year (per 1000 population using World Bank data) in the origin countries. Results: A total of 135 888 physicians emigrated from WHO-designated shortage countries during 2000 to 2021 and 516 030 physicians emigrated from nonshortage countries. Compared with nonshortage countries, there was a decrease in physician outflow by 47.03 physicians (95% CI, −92.29 to −1.76 physicians) per country per year after 2010 in WHO-designated shortage countries and the WHO Code was associated with a reduction of nearly 30% in the average annual outflow of physicians from these countries. However, there was a slight decrease in physician density in the WHO-designated shortage countries after 2010 (−0.22 [95% CI, −0.33 to −0.11] physicians per 1000 population) compared with the nonshortage countries. Conclusions and Relevance: This study found that voluntary implementation of the WHO Code was associated with lower physician outflow from WHO-designated shortage countries without improvement in physician density in those countries.

This cohort study explores the likelihood of local childbirths in US counties before and after the 2009 expansion of the maternal health clinician workforce via the National Health Service Corps.