AI Health
Friday Roundup
The AI Health Friday Roundup highlights the week’s news and publications related to artificial intelligence, data science, public health, and clinical research.
May 15, 2026
In this week’s Duke AI Health Friday Roundup: vibe coding in research; frameworks for AI use in healthcare; tracking the diffusion of bogus citations in the literature; reconsidering frameworks for AI regulatory approval; the existential doubt being sown by AI; immunotherapy trial for glioblastoma reports phase 1 results; personal information being surfaced by AI search; deciphering the noncoding genome; much more:
AI, STATISTICS & DATA SCIENCE
- “At its purest, vibe coding doesn’t involve looking at the code — just the product. But the term has no strict definition, so what counts as vibe coding is fuzzy. Plenty of people with coding know-how start a project by vibing and then check the code by hand, or start coding by themselves and then ask an AI tool to fill in the gaps.” A Nature news feature by Nicola Jones walks readers through the process of using AI-assisted “vibe coding” to assist in research.
- “The fabricated references we identified were not obviously defective: topically specific, correctly formatted, attributed to real researchers, and beared [sic] plausible publication dates. Systematic reviews have found that approximately one in four references in medical journal articles contains errors, confirming that reference verification is not standard in peer review. Automated reference verification can close this gap.” An article published in Lancet by Topaz and colleagues presents findings from a survey of scholarly biomedical publications that shows a steep increase in the incidence of fabricated citations in recent years.
- “Experts say that these privacy lapses are most likely due to personally identifiable information (PII) being used in training data, though it’s hard to understand the exact mechanism causing real phone numbers to show up in the AI-generated responses. But no matter the reason, the result is not fun for people on the receiving end—and, even more worryingly, there appears to be little that anyone can do to stop it.” At MIT Technology Review, Eileen Guo reports on the apparent doxxing of people by Google AI, which has been revealing their personal contact information, sometimes in connection with entirely unrelated queries.
- “We find a sharp rise in non-existent references following widespread LLM adoption, with a conservative estimate of 146,932 hallucinated citations in 2025 alone. These errors are diffusely embedded across many papers but especially pronounced in fields with rapid AI uptake, in manuscripts with linguistic signatures of AI-assisted writing, and among small and early-career author teams. At the same time, hallucinated references disproportionately assign credit to already prominent and male scholars, suggesting that LLM-generated errors may reinforce existing inequities in scientific recognition.” A preprint by Zhao and colleagues, available from arXiv, reports on a study examining the extent of publication citations “hallucinated” by LLMs and included in published papers.
- “The rapid integration of AI/ML into clinical practice demands that physicians understand what FDA clearance does and does not guarantee. Strengthening oversight will require coordinated action across regulatory agencies, Congress, health systems, and device manufacturers, but these reforms would give clinicians the information and infrastructure they need to deploy AI tools responsibly.” A perspective article published in the Annals of Internal Medicine by Rosen and Mandl criticizes the FDA’s current standards for regulating health AI under existing medical device frameworks and calls for new approaches.
BASIC SCIENCE, CLINICAL RESEARCH & PUBLIC HEALTH
- “As AI systems become more embedded in care delivery, these challenges pose significant risks to procurement efficiency, patient safety, and quality of care. Vendors often face the challenge of safeguarding proprietary information while responding to the evolving expectations of HDOs. The framework helps navigate this tension by highlighting what types of information HDOs find most useful…” A perspective article published by Kpodzro and colleagues from the Health AI Partnership in NEJM AI proposes a framework for ensuring transparency around the adoption of AI tools in patient care settings.
- “Thunderstorms have captivated humanity for millennia, and yet their inner workings remain deeply mysterious. Storm clouds are opaque. They’re dangerous to approach. And they’re too big to fit in a lab. Inquisitive researchers have been sending kites, balloons, and rockets up into them for nearly three centuries, and they’ve learned a lot. But every time lightning lovers get closer to the action, they discover major gaps in their understanding.” Quanta’s Charlie Wood takes a look at the surprisingly complex field of study that’s trying to elucidate why and how lightning happens.
- “We have shown that personalized neoantigen DNA vaccines, using the GNOS-PV01 platform, are safe and well-tolerated in the post-radiotherapy adjuvant setting….this study demonstrates that personalized neoantigen DNA-based vaccines can provide a potential adjuvant therapeutic option for patients with GBM [glioblastoma], with encouraging outcomes as demonstrated by one-third of patients surviving 24 months or longer.” A report on a phase 1 study of an immunotherapy vaccine for the brain cancer glioblastoma, published this week in Nature by Garfinkle and colleagues, shows promising findings.
- “…researchers are making progress on decoding the regulatory region of the genome, and learning the underlying grammar of the elements that govern when and where genes are turned on and off. To do so, they have turned to a suite of methods known as massively parallel reporter assays (MPRAs).” Nature’s Michael Eisenstein reports on recent advances in understanding the large expanse of the non-coding portions of the human genome.
COMMUNICATIONS & Policy
- “… one of the great things about writing and the internet in general is that there are all sorts of different dialects and styles and things that can work online. And so maybe what I have been noticing is a sameness, a homogenizing of large parts of the internet, including places I often felt were very human. This is objectively happening, researchers believe.” In a personal perspective for 404, Jason Koebler confronts the hall of mirrors raised by the constant possibility that the words and images you’re seeing are generated at least in part by AI.
- “The rapid integration of AI into health care presents a troubling paradox. The same tools that promise to democratize medical knowledge may instead deepen the divide between well-resourced health systems and safety-net hospitals. How we govern AI adoption in the coming years will determine whether it becomes medicine’s great equalizer or its newest social determinant of health.” In a perspective article for NEJM AI, Sam Joseph Amirfar explains why the decisions being made now around the adoption of AI in healthcare will reverberate for patients – for good or ill.
- “Five of the country’s largest book and journal publishers along with bestselling author Scott Turow have joined together to file a class action lawsuit against Meta and its founder and CEO, Mark Zuckerberg, for willful infringement of millions of works, torrented via pirate sites, to develop Meta’s Llama large language models.” Publisher’s Weekly reports on a lawsuit being brought against Meta by publishers and authors, who accuse the tech giant of having violated copyright in the training of the company’s large language model.
- “If we are unimpressed by stories about paperclip maximizers remaking the galaxy, omniscient bureaucracies of terror or wonder, markets that suddenly become self-aware, and the like, it is not because we think they are too weird. Rather, they are not nearly weird enough, and miss how much of the weirdness is already here. The possible futures we face are much messier and more varied than stark visions of omnipotent AGI, just as our immediate past was.” An essay posted by Henry Farrell and Cosma Rohilla Shalizi at Columbia’s Knight Institute presents a skeptical critique of the idea that AI will inevitably transform human society, at least in the ways many tech leaders suggest.
