RISEUPTOHIV
RISEUPTOHIV
EDUCATION • COMMUNITY • RESOURCES
About Contact
ES HT
Science & Technology — Learning Hub

AI in HIV Healthcare:
What It Can and Can't Do

Last reviewed:

Health information — Educational information, not medical advice. Talk with your healthcare provider about your care and treatment decisions.

AI is being used and studied in HIV communications, prevention services, and research. Here is what the evidence supports, what remains uncertain, and why human responsibility and community experience belong at the center.

Share this article Facebook X WhatsApp Text Email LinkedIn Reddit Threads Bluesky
← Learning Hub

Artificial intelligence is already part of some HIV communications and service programs, while other applications remain research projects or proposals. These are different stages of evidence. A tool that predicts an outcome is not automatically proven to improve it, and a study or funding announcement does not establish routine clinical use. This article separates those stages and links to the organizations and studies behind each example.

This page starts with transparency about this site, because that's the right place to start. Then it covers what AI is doing across the HIV field, what the risks are, and how to engage with AI-generated health information thoughtfully and safely.

How AI helped build RiseUpToHIV

I’m Kevin — a person living with HIV for 16+ years, based in Tampa Bay, FL. I have lived with HIV since March 2010, and I brought years of advocacy, patient education, and community experience to this project. Over eight months and 800 hours, I used Perplexity, Claude, and ChatGPT as my primary tools for research, writing, and web development, working from my home office in Tampa Bay, Florida. I built the project as one person, with a breadth often associated with a larger, well-funded team. The Florida locator became an early focus in March 2026.

I retained full editorial control. I chose every topic, the site and hub structure, the Learning Hub, and every color and visual choice. I repeatedly requested edits and re-checks, changed wording, and rejected material. Like an electrician’s tools, AI helped me carry out the work; the plan, choices, and editorial responsibility remained mine.

Checking AI output is work: reading the original source, checking its date and context, following a citation, and revisiting a claim when it does not hold up. Agreement between models is not independent verification. Lived experience adds context; it does not replace clinical expertise or evidence.

I am responsible for this site and its corrections. This disclosure does not mean every page has received an independent clinical review. If you find something inaccurate or unclear, please report it. Read the full disclosure or my story.

Using AI for good, while taking concerns seriously

Site disclosure and public-opinion context updated September 26, 2026. The clinical and research sections retain their September 20 review date.

I understand why troubling stories about AI can leave people uneasy. Concerns about misinformation, privacy, bias, and its effects on people’s work deserve a serious response. In a June 22–28, 2026 survey, Pew Research Center found that 52% of U.S. adults felt more concerned than excited about the growing use of AI in daily life; 9% felt more excited and 37% felt equally concerned and excited.[23] I respect that unease, including the views of people who choose not to use these tools.

I built RiseUpToHIV as an example of using AI for good: helping one person bring HIV education, care resources, history, and community voices together in a free, accessible place. Education, community, and resources are the site’s main purpose; its policy coverage supports understanding and self-advocacy, with links to organizations leading broader campaigns. AI was a primary tool for research, writing, and web development. My lived experience, decisions, and repeated requests for revisions shaped the project. The purpose is to help people understand their options, find support, and feel less alone.

Responsible use requires continuing work: following sources, checking dates and context, revising unsupported wording, explaining limitations, and responding to corrections. Agreement between AI tools does not verify a claim. This website remains an educational resource, and it does not claim independent clinical review of every page. I am responsible for the work and for improving it when something is wrong.

For me, the opportunity is personal. HIV information reaches people through news stories, clinic conversations, social posts, research papers, newsletters, and the daily work of community organizations. RiseUpToHIV is my attempt to connect some of those paths so someone can move from a question to an explanation and a practical next step.

I built it while grieving and living with health limitations, often from a desk at home. In 2026, 45 years after the first reported AIDS cases in 1981, the knowledge available to me exists because people living with HIV, activists, allies, researchers, scientists, and care teams kept doing the work. Many did not live to see what they made possible.

For me, this launch marks a meaningful meeting of that history with new tools: the opportunity for an individual advocate with lived experience and editorial responsibility to bring information together on a scale I could not previously have managed. That is why this moment matters to me, and why I want the work to honor the people behind it.

Productivity should be judged by the usefulness and care in the result, including the human time spent checking and correcting it. A faster draft does not tell us whether someone understood the information, trusted it for good reasons, or found the help they needed. Those are the outcomes I want to learn from.

I would welcome the chance to contribute. I am interested in volunteering with working groups on responsible AI in HIV education and sharing practical lessons with community organizations. I offer a lived-experience perspective and what I have learned while building this site. Get in touch about that work →

What's happening: AI across the HIV field

There are several kinds of AI in this field: models that find patterns in health data, systems that help interpret images, and generative tools that produce text or conversation. HIV.gov documented its use of generative AI for communications in December 2023, with staff review and disclosure.[3] The examples below also include an implemented PrEP messaging system, a community-linked self-testing study, and an announced cure-research project. Their results and limitations should be evaluated separately.

A broad public-health example is not automatically an HIV-specific example. CDC’s current AI overview describes syndromic surveillance and influenza forecasting; it does not establish that CDC routinely uses AI to forecast HIV outbreaks.[1] Claims about a particular HIV program need evidence about that program.

A documented international discussion. Michael J. A. Reid and colleagues announced the Lancet Global Health Commission on artificial intelligence and HIV in April 2025.[4] The announcement establishes a dedicated initiative on equitable and sustainable AI use in the HIV response; it should not be described as a completed clinical guideline.

Guidance is being written in real time — and community voices belong in that room

HIV-specific resources and ethical recommendations have already been published. AVAC is an active contributor alongside Audere, ITPC, Georgetown’s O’Neill Institute, WHO, and others. The sources do not establish a single organization as the leader of the entire field. An advocates’ brief, an ethics report, a conference discussion, and a clinical guideline also serve different purposes.

AVAC and Audere — a published guide, not just a document in development. AI and HIV Programs: A Guide for Advocates was published on November 21, 2025. The two-page brief introduces possible applications and calls for community involvement, data safeguards, and evaluation as projects move from pilots toward wider use.[10] It is an advocacy resource, not a clinical practice guideline or approval of particular AI products. AVAC’s AI & HIV newsletter began with a May 2026 issue.[21]

Georgetown / O’Neill Institute — published ethics recommendations. On October 14, 2025, the institute published Recommendations on the Ethical Use of Novel HIV Data & Analytics. An 18-member international working group, including people living with HIV, developed guidance for researchers, program designers, funders, and multilateral organizations.[11] This is direct evidence that documented working-group discussions and HIV-specific recommendations already exist.

UNAIDS Global AIDS Strategy 2026–2031. The strategy includes digital health and AI among the priorities for the HIV response, with attention to ethics and community participation.[12] A strategic commitment describes a direction for work; it does not by itself establish that a particular tool is safe or effective.

AIDS 2026 — a documented conference agenda. AVAC’s conference roadmap lists “The digital care continuum: Connecting AI, PrEP, and people” for July 30, 2026.[13] The roadmap verifies a scheduled discussion. It does not, by itself, identify AVAC as the session organizer or prove that the tools discussed improve health outcomes.

WHO — a recorded HIV and AI webinar. WHO’s South-East Asia office reports that its May 26, 2026 webinar brought together more than 150 participants, including HIV program managers, civil society, and implementing partners. The event page provides recordings and presentations on digital health and AI in HIV programs.[14] ITPC’s March 20 report describes its March 16, 2026 community-focused webinar on AI, HIV, and governance.[22]

Where RiseUpToHIV fits in. This is an independent education project learning from published guidance and practical experience. My interest is in contributing to that work, not claiming a formal role in these organizations or their working groups. Community participation should shape a project from its first questions through evaluation, correction, and decisions about whether it should continue.

Who is using or studying AI in the HIV response

Communications · documented organizational use
HIV.gov: creating information with human review
HIV.gov describes using generative AI in communications, including blog, social, and video work, with disclosure and staff quality checks.[3] That documents an editorial workflow, not an autonomous clinical advice service. Its 2023 account also records training labs and a workshop for colleagues.
PrEP services · implemented, observational evidence
AHF and Healthvana: supporting PrEP engagement
A 2026 study examined Healthvana’s conversational AI at AIDS Healthcare Foundation clinics using records from 155,217 eligible adults. People who interacted with it had higher PrEP initiation and follow-up rates.[16] The study was observational: users chose whether to engage, so the findings do not prove AI caused the difference or reduced HIV incidence.
Prediction · model development and validation
Researchers: identifying potential PrEP candidates
Krakower and colleagues developed and validated a prediction model using electronic health records to help identify potential PrEP candidates.[2] That is evidence about model performance. It does not show that the model diagnoses HIV, replaces a patient-clinician conversation, or improves outcomes whenever it is deployed.
Cure research · announced research project
amfAR: the HIV Immune Atlas Study
In April 2026, amfAR announced a $2 million collaboration involving Gladstone Institutes, Weill Cornell Medicine, Mount Sinai, and Helmholtz Munich. The project plans to combine single-cell data and AI to study the HIV reservoir.[18] It is a research effort intended to support future discoveries; it is not an AI-developed cure or a clinical treatment.
Community services · feasibility and acceptability study
Audere and Shout-It-Now: Aimee
Audere’s study materials describe a WhatsApp AI companion linked to Shout-It-Now services for adolescent girls and young women in South Africa. The toolkit includes HIV information, self-testing support, computer-vision test interpretation, nurse review, and referral options.[17] These materials establish the study’s design, not a demonstrated reduction in HIV acquisition.
Public health · proposed applications
CDC-authored research: possibilities for prevention
A 2025 paper by CDC and Emory authors examines possible AI applications in HIV prevention, including surveillance and care engagement.[5] It reports that a January 2025 search found no AI interventions in CDC’s HIV prevention compendium. That dated finding is not a claim about today’s entire field. The paper does not establish routine AI deployment in Data to Care.

NIH’s role includes supporting research. NIH’s archived funding call, Multimodal Artificial Intelligence to Accelerate HIV Clinical Care, sought development and evaluation of AI models for HIV applications, with ethical safeguards and stakeholder engagement.[19] A funding call documents a research priority; it is not proof that NIH has deployed a finished clinical tool.

A 2021 qualitative study interviewed 40 UK participants in 2020 about sexual-health chatbots. Some valued convenient, nonjudgmental information; others questioned the tools’ content, trustworthiness, and accuracy. These interviews explored perceptions, not improved health outcomes or a guarantee of privacy.[15]

The real risks: what can go wrong

Hallucinations
AI language models generate confident-sounding text that is sometimes factually wrong. In healthcare, a hallucinated drug dosage, interaction warning, or treatment recommendation can cause direct harm. See the hallucinations section below for detail.
Racial and demographic bias
AI systems can reproduce or amplify bias in their data and design. Obermeyer and colleagues documented racial bias in a health-management algorithm that used healthcare spending as a proxy for illness. This was not an HIV-specific study; it shows why HIV programs need to evaluate their own tools and outcome measures.[7]
Privacy and data security
Sharing HIV status, symptoms, or medication information with AI chatbots raises real privacy questions. Data retention, human access, and training practices vary by service, account type, and settings. Do not assume a conversation has the protections of a clinical encounter. HIV status is among the most sensitive categories of health information.
Outdated information
A model’s training data has a cutoff; some tools can retrieve newer information, but retrieval does not guarantee that a response is current or correct. HIV treatment guidelines change. New drugs are approved. Funding situations shift. An AI trained on data from 2023 may give you outdated information about a topic that has changed significantly since then.
No physical examination
A general-purpose chatbot cannot perform a clinician’s physical examination or reliably assess your complete medical situation. Image or audio features do not make it equivalent to a clinical visit. Generalized AI advice cannot substitute for individualized clinical assessment.
Misinformation amplification
AI can produce HIV misinformation — including HIV stigma, outdated information about transmission risk, or inaccurate statements about treatment — with the same confident, authoritative tone it uses for accurate information. Tone and presentation alone cannot reliably distinguish a correct statement from a fabricated one.[20]
The appearance of expertise
AI responses sound like they were written by a knowledgeable expert. Fluent, well-organized, confident prose reads as authority whether or not the substance behind it holds up. That credibility is not earned — it is a feature of how the technology writes, not an indicator of accuracy.
Lack of accountability
A chatbot is not a licensed clinician. Before a health organization adopts an AI system, it should identify who is responsible for reviewing outputs, responding to harm, correcting errors, and handling complaints. Accountability should not disappear behind the technology.

Hallucinations: the specific risk for HIV information

In this context, a hallucination is an AI-generated statement or citation that appears plausible but is unsupported or wrong. A model can produce such an error even when a tool can retrieve outside sources; generating fluent text is different from verifying a claim.[20]

For general topics, hallucinations are annoying. For medical topics — and especially for HIV — they can be dangerous. Examples of the kind of hallucinations that can occur in HIV contexts include: incorrect drug interaction warnings (or missing real ones), wrong information about PrEP dosing schedules, outdated information about HIV criminalization laws, inaccurate statements about transmission risk that could lead to unsafe decisions, and fabricated statistics that sound plausible but are wrong.

A response can mix accurate information with an important error. For example, an otherwise useful explanation could contain an incorrect dosing instruction. That is an illustrative risk, not a measured claim that any fixed percentage of an answer is accurate.[20]

A small study illustrates why checking matters. A 2026 paper evaluated answers to 26 HIV questions from models identified by its authors as ChatGPT-3.5 and DeepSeek-R1. Three clinicians assessed the responses; the authors reported generally high scores alongside some misleading content.[6] This was a limited assessment of particular model outputs, not a clinical trial or proof that current chatbots are safe for individual treatment decisions.

Equity and bias: the HIV-specific concern

The equity dimensions of AI in HIV healthcare deserve specific attention — because the populations most affected by HIV are the same populations most at risk of being harmed by AI bias.

Training data and design choices. A model can perform unevenly when the people or circumstances it encounters differ from its development data. A 2019 study found racial bias in a widely used health-management algorithm that used spending as a proxy for health needs.[7] That example was not specific to HIV. It illustrates why HIV programs should test their own tools across relevant populations and examine what the model is actually measuring.

Inclusive language requires review. A 2024 study involving clinicians, HIV community members, and gender and queer studies experts found that ChatGPT’s HIV prevention answers could be scientifically accurate while still lacking context and inclusivity.[8] The findings support involving community reviewers alongside clinical reviewers. Accuracy and respectful, useful communication both matter.

Who can use the tool? Internet access, language, disability access, trust, and staff capacity can determine who benefits from an AI service. Programs should evaluate these barriers locally rather than assume an app reaches everyone. Community involvement should include people who cannot or do not want to use the technology.

Responsible use requires people to check the technology, protect privacy, and involve the communities affected.
A summary of the concerns discussed by Kamitani, Koenig, and Sullivan in AIDS, 2025.[5]

What ethical AI use in HIV looks like

The following are the commitments I aim to put into practice in this project. They are a community-informed editorial approach, not a certification, a clinical guideline, or a claim that the work is finished.

1
Disclose AI use transparently
People consuming health information deserve to know when AI was involved in creating it. Transparency about AI use is not optional — it is the baseline of trust. This site discloses AI use on its Disclaimer page and here. We believe every health organization using AI should do the same.
2
Human expertise and lived experience must drive the work
People who understand the topic and the community must direct the work. Lived experience, professional or volunteer experience, and relevant clinical expertise bring different kinds of knowledge. Be clear about who is responsible, what they know, and when further expert review is needed.
3
Never replace clinical care
AI-generated health information — however accurate and well-sourced — cannot replace individualized clinical assessment. Any AI tool used in HIV contexts should be explicit about this limit, should direct people to appropriate care when clinical questions arise, and should never represent itself as equivalent to medical advice.
4
Verify against trusted sources
Check clinical claims, statistics, and recommendations against the original research or appropriate current guidance. Read beyond the headline, confirm that the source supports the actual wording, and explain uncertainty. A plausible citation, a second model’s agreement, or an automated link check is not proof of accuracy. Keep a route for readers to report problems and for corrections to be made.
5
Protect privacy — never use sensitive disclosures for AI training
HIV status, symptoms, treatment history, and related personal information should never be shared with AI tools that use conversation data for training. The site does not offer an AI chat assistant. Its form and moderation services, and the privacy limits of public contributions, are explained in the Privacy Policy.
6
Center equity — who isn't being reached?
AI tools in HIV should be evaluated not just by how well they work for typical users, but by how well they serve the communities most affected by the epidemic: Black and Latino communities, rural Southerners, transgender people, people who inject drugs, undocumented immigrants. If an AI tool serves English-speaking insured urban patients well but fails everyone else, it is not a solution to the HIV epidemic — it is a tool that benefits the already advantaged.

For patients: how to use AI health information safely

If you use a general-purpose AI tool to look up HIV information, keep these limits in mind.

Use explanations as a starting point. An AI tool may help explain a term or organize questions for an appointment, but it can also get basic facts wrong. It should not determine your medication, dose, testing schedule, or treatment plan.

Verify clinical information before acting. Check claims about medications, lab results, and treatment options against current guidance and a qualified healthcare professional. This website also needs checking and correction; it is not an independent clinical authority. Use appropriate official or professional sources for legal questions.

Protect personal information. Avoid entering identifying health details into general-purpose tools. Privacy, retention, and training practices differ by service and settings, and turning off model training does not eliminate every privacy risk. Discuss your individual care through appropriate clinical channels.

Ask AI to cite sources. When using an AI tool for health information, ask it where the information comes from. A well-designed AI should be able to point you toward CDC, NIH, or peer-reviewed research for factual claims. If it can't — or if it fabricates citations — that's a signal to verify more carefully.

Do not use tone as an accuracy test. A confident answer can be wrong, and cautious wording alone does not make an answer reliable. Check the supporting source and consult a qualified professional for decisions about your care.[20]

The practical test: Can you open a reliable, current source that supports the claim, and does the advice need a clinician who knows your situation? A confident answer or a convincing citation is not enough.

The future: where AI and HIV are heading

The trajectory of AI in HIV healthcare over the next five to ten years will be shaped by how successfully the field navigates the tensions between AI's genuine potential and its real risks.

Possible future applications include helping people navigate prevention services, identifying gaps in care, and supporting research. Those possibilities require evaluation in the settings where they would be used. A system should not be assumed to prevent outbreaks, improve retention, or eliminate disparities simply because it uses AI.

Some risks can be reduced through evaluation, oversight, privacy protections, and community participation; those measures do not guarantee safety. The HIV community has demanded meaningful participation for more than four decades. The Denver Principles[9]—the 1983 statement asserting the rights and participation of people with AIDS—offer an enduring foundation for asking who makes decisions and whose knowledge counts.

The evidence will keep changing. Programs should be willing to revise or stop a tool when it is unsafe, inequitable, or not useful. Community feedback and human responsibility belong throughout that process.

RiseUpToHIV is one person’s attempt to put these principles into practice: a personally funded resource with 200+ pages, shaped by lived experience, community knowledge, and AI-assisted work under human editorial control. Its value will depend on whether people find it useful and how carefully I listen, check, and correct it. I hope it can help more people reach the knowledge and support this community has built.

References & Sources

  1. CDC, How CDC is Using AI — March 13, 2026; general public-health examples, not verification of routine HIV outbreak forecasting. ↩
  2. Krakower DS et al. Development and validation of an automated HIV prediction algorithm to identify candidates for pre-exposure prophylaxis: a modelling study. Lancet HIV. 2019;6:e696–e704. Development cohort: 1,155,966 Atrius patients; AUC 0.86 in development, 0.91 in prospective validation, and 0.77 in external validation Source. ↩
  3. HIV.gov, "HIV.gov’s Use of Generative AI" (December 14, 2023) — HIV.gov’s own account of how it uses generative AI, including watermarks, the disclaimer "Developed by AI, reviewed by Humans," and team quality assurance on every AI-assisted piece. ↩
  4. Reid MJA, Otieno BA, van Heerden A, et al., Announcing the Lancet Global Health Commission on artificial intelligence (AI) and HIV: leveraging AI for equitable and sustainable impact — Lancet Global Health, April 2025; commission announcement. ↩
  5. Kamitani E, Koenig LJ, Sullivan P, Transformative Potential of Artificial Intelligence in US CDC HIV Interventions: Balancing innovation with health privacy — AIDS, 2025; discussion of potential applications, with a January 2025 compendium search. ↩
  6. Huang H, Zhang H, Qin X, et al., How AI responds to common HIV/AIDS questions: ChatGPT versus DeepSeek — Frontiers in Public Health, August 2026; a small evaluation of model responses. ↩
  7. Obermeyer Z, Powers B, Vogeli C, Mullainathan S. Dissecting racial bias in an algorithm used to manage the health of populations. Science. 2019;366:447–453; doi:10.1126/science.aax2342 Source. ↩
  8. De Vito A, Colpani A, Moi G, et al., Assessing ChatGPT’s Potential in HIV Prevention Communication: A Comprehensive Evaluation of Accuracy, Completeness, and Inclusivity — AIDS and Behavior, 2024; author-institution record and full-text link. ↩
  9. Advisory Committee of People with AIDS. The Denver Principles (1983). Original statement on self-determination, participation and rights; reproduced by UNAIDS Source. ↩
  10. AVAC & Audere, AI and HIV Programs: A Guide for Advocates — November 21, 2025; two-page advocacy brief, with downloadable PDF. ↩
  11. O'Neill Institute for National and Global Health Law (Georgetown University), "Recommendations on the Ethical Use of Novel HIV Data & Analytics" (October 2025) — multi-part guidance from an 18-member international, interdisciplinary working group supported by the Bill & Melinda Gates Foundation. ↩
  12. UNAIDS, "The Global AIDS Strategy 2026–2031" — explicitly names AI and digital health as a strategic priority requiring clear ethical and technical guidance, national implementation support, and strong community engagement. ↩
  13. AVAC, AIDS 2026 Conference Roadmap — Lists session SAT45 on AI, PrEP, and people for July 30, 2026; agenda evidence only. ↩
  14. WHO South-East Asia, Integrating Digital Health and Artificial Intelligence into HIV Responses: Lessons, Innovations and Opportunities — May 26, 2026; event report, recordings, and presentations. ↩
  15. Nadarzynski T, Puentes V, Pawlak I, et al. Barriers and facilitators to engagement with artificial intelligence (AI)-based chatbots for sexual and reproductive health advice: a qualitative analysis. Sexual Health. 2021;18(5):385–393; doi:10.1071/SH21123. Forty UK participants were interviewed in January–June 2020 Source. ↩
  16. Narayan A, Blasingame M, Warmuth S, et al., AI-augmented communication improves HIV PrEP initiation and persistence in populations disproportionately impacted by HIV — npj Digital Medicine, 2026; observational study of Healthvana at AHF clinics, not randomized causal evidence. ↩
  17. Audere, Aimee: Study Participant Information and Informed Consent — Feasibility and acceptability study with Shout-It-Now; protocol and participant information. ↩
  18. amfAR, amfAR Launches $2 Million AI-Enabled, Collaborative Study to Map the HIV Reservoir — April 8, 2026; research announcement. ↩
  19. NIH / NIMH, Multimodal Artificial Intelligence to Accelerate HIV Clinical Care (R01 Clinical Trial Optional) — Archived RFA-MH-26-105; research funding call, now expired. ↩
  20. CDC. Considerations for Generative AI in Public Health, March 13, 2026. Source. Checked September 20, 2026. ↩
  21. AVAC. AI & HIV resource page and newsletter archive, beginning May 2026. Source. Checked September 20, 2026. ↩
  22. ITPC Global. Are the Robots Taking Over? AI, HIV, and the Fight to Keep Communities in Charge, March 20, 2026. Source. Checked September 20, 2026. ↩
  23. McClain C, Park E. Young adults in the U.S. are increasingly wary of AI, concerned it will take jobs. Pew Research Center, August 18, 2026. National survey of 3,488 U.S. adults conducted June 22–28, 2026; reports overall concern and excitement about AI. ↩
Search RiseUpToHIV