Editorial Note: At AIDS 2026, Sarah Morris, Chief Product Officer of Audere, delivered a presentation titled From Algorithms to Empathy: Artificial Intelligence in Self-Care Support for HIV Services, sharing cutting-edge practical experience and exploring the value and boundaries of AI technology deployed in long-term HIV health management. As AI gradually penetrates multiple self-care scenarios including medication adherence tracking, psychological counseling and risk monitoring for people living with HIV, balancing technical efficiency with humanistic care, data security, and equitable universal access have emerged as core issues demanding clarification across the industry.To this end, Infectious Disease Frontier conducted an exclusive interview with Natalie Maricich, Product Lead for Southern Africa, from the Audere team. Drawing on frontline practical outcomes, the in-depth dialogue covers the current landscape of AI promotion, technical limitations, and the development of long-term regulatory standards. It elaborates on the team’s insights into co-creation-driven R&D, security governance and ethical framework construction, offering international reference ideas for the domestic application of AI in long-term HIV care.
Infectious Disease Frontier: Many current AI tools focus primarily on data monitoring and risk alerts. In the context of HIV healthcare services, how can we balance technical algorithmic capabilities with the need for human empathy if AI is to effectively support self-care among people living with HIV?
Natalie Maricich: Empathy starts with co-design, not model development. We do not build technology for people living with HIV—we co-develop solutions with them, and this distinction is critical. Health challenges extend far beyond clinical concerns and entail a host of real-world issues requiring wraparound support. Therefore, the true value of AI in this context lies not merely in tracking data and pushing risk reminders, but in establishing a trustworthy, consistent interaction channel that motivates sustained user engagement. It assists individuals in decision-making across diverse dimensions and delivers tangible support to address their multifaceted real-life struggles.
I wish to emphasize that balancing technical capacity and human empathy carries profound significance. Empathy is not an afterthought activated once technical limits are reached. We must clearly define the applicable scenarios and inherent limitations of AI, and accurately identify junctures requiring human intervention—this stands as an indispensable core principle.
Infectious Disease Frontier: AI-enabled HIV self-care covers multiple areas including medication adherence management, psychological support and symptom monitoring. Based on your practical research, how effective has the real-world rollout of such AI applications been at this stage? What notable limitations exist in terms of technology or user acceptance?
Natalie Maricich: Our suite of AI-assisted care tools has moved beyond pilot testing and delivered measurable real-world impacts. In our initial AIMEE study, we found that 45% of users who engaged with the AI companion successfully connected with professional medical services, leading to measurable improvements in their health status. Moreover, for marginalized and vulnerable populations, this anonymous, non-judgmental service gateway often represents their sole viable access point to health information and care resources.
Objectively speaking, notable limitations are centered around privacy, data security, and user trust. On the cognitive front, general users may harbor misgivings about AI reliability, while clinical practitioners fear AI may displace human healthcare roles. On a practical level, digital inequity—driven by uneven access to devices and internet connectivity—risks perpetuating or even widening existing gaps in healthcare resource distribution. These are all valid concerns that cannot be overlooked.
We uphold a foundational guiding principle: AI is designed to augment and support clinical staff, never to replace human-led medical care. Ultimately, the industry’s central challenge lies in building AI products that earn user trust while rigorously demonstrating clinical and service value. For this reason, ethical co-design is not an optional add-on for the sector; it is an essential professional obligation. We must maintain continuous iterative refinement aligned with evolving technological advancements and industry standards, consistently advancing both ethical and technical frontiers to ensure AI tools operate compliantly, securely and sustainably.
Infectious Disease Frontier: As AI becomes increasingly involved in HIV health management, issues such as data privacy, algorithmic bias and medical liability have attracted considerable attention. For future scale-up, what regulatory and normative framework do you believe should be established to ensure safe and sustainable integration of AI into the long-term self-care system for people living with HIV?
Natalie Maricich: Safeguards for AI medical applications rest on two core dimensions: first, data security, centered on comprehensive protection of sensitive personal health data and privacy belonging to vulnerable groups; second, application safety, mitigating potential harm stemming from the technology itself. Within the highly private care context of HIV services, erroneous guidance output by AI, insensitive communication tones, or rigid one-size-fits-all standardized service models can exert profound adverse impacts on users. For this reason, context-specific and localized adaptation tailored to HIV care workflows is an indispensable prerequisite for on-the-ground technical deployment.
Concurrently, we have built a proprietary multi-layered comprehensive safety governance framework that underpins all real-world rollouts. This hybrid framework combines automated evaluation for accuracy and stigma with a closed-loop human review mechanism; all AI medical tools must implement this dual-audit workflow before public deployment. Additionally, equity constitutes a non-negotiable tenet of regulatory design: the people who most need support are sometimes hardest to reach digitally, so AI has to be one channel among several, not a replacement for in-person and community health worker services.
On the regulatory front, I advocate for mandatory mandates: all AI tools must complete localized adaptation calibration and full-spectrum safety testing prior to large-scale rollout. Amid widespread industry budget cuts for HIV prevention and treatment, standardized frameworks must also embed long-term sustainable operation planning into the initial project design phase, rather than treating sustainability as a post-hoc amendment. This approach guarantees that AI-enabled HIV self-care models can operate compliantly and sustainably into the future.
