When Cassie Ridgway held her first clothing swap in Portland, Oregon, 14 years ago, she had a few goals: keep clothes out of landfills, help people find free fashion treasures and build community.
The swap attracted about 150 people, and grew from there. Now, the twice-yearly event, which organizers call The Biggest Swap in the Northwest, draws between 500 and 850 participants to share clothes and accessories in a partylike atmosphere.
“We have a DJ and two full bars, so there’s some singing and dancing. But no one’s getting drunk at 1 p.m. on a Sunday afternoon,” said Ridgway’s co-founder, Elizabeth Mollo.
The swap is part of a larger movement across the country to share resources with neighbors — one shirt, meal or book at time.
The Portland event asks for a $10 entry fee to cover costs, but the clothes are free and there’s no limit to how much participants can take. People bring their gently used clothing, shoes and accessories to a sorting station, where volunteers sort it into bins and onto tables.
Ridgway, who worked in the apparel industry, sees the process as an answer to throwaway “fast fashion.” She describes “the ‘peak pile’ moment, when our sorters are summiting a mountain, a literal tonnage of apparel, sorting as quickly as they can. In this moment, we see the true ramifications of consumer culture and waste.”
Leftover clothing is donated to another free neighborhood swapping event.
Ridgway recalls a single mom telling her she was able to outfit her teenager with Nike shoes and other major brands typically outside her price range. “These conversations, and so many others, have truly kept me coming back to this event,” she says.
There are no dressing rooms, so participants are encouraged to come in tight-fitting clothes and try things on where they are.
“It does get a little chaotic,” Mollo says, but many people return year after year.
“Where else can you get a whole new wardrobe for $10?”
As prices climb for many food items, community resource-sharing becomes increasingly important, says Taylor Scott in Richmond, Virginia.
Scott was a recent college graduate when the pandemic put her dream of becoming an FBI agent on hold. She took up gardening, and quickly found herself with more tomatoes than she could consume. A friend suggested she put the extras into a community refrigerator, like ones they knew of in places like New York City. But Scott found there was nothing of the sort in Richmond.
“I decided that was what I was going to do for my birthday,” she says.
Scott hopped on Instagram to see if her friends wanted to help, and quickly received an offer of a fridge and a promise to paint it. Several months and planning calls later, she opened her first community fridge outside a cafe, in January 2021.
It was a hit.
“Right away, people asked me when I was going to open more,” Scott says.
She built relationships across the city on “word of mouth and faith” as she added fridges over the next four years. As the project grew and became RVA Community Fridges, food donations expanded from restaurants and farms to include private events and weddings.
“We’ve saved so much food that would have gone to waste,” Scott says.
Today, the 27-year-old president of RVA Community Fridges and her crew of volunteers run 14 fridges, offer “farm to table” education classes and hold community cooking days at a kitchen. The organization has given away more than 520,000 pounds of food, Scott says.
She also likes that the fridge sites have become neighborhood gathering spots. She’s seen people who once needed the food share become volunteers when they’re in a better place.
“They started out taking and now they’re giving,” Scott says.
This style of hyper-local sharing is also a hallmark of Little Free Library, the nonprofit behind those cute little book huts that dot communities nationwide. The libraries offer round-the-clock access to free books, and are meant to inspire meaningful interactions.
“People tell me they’ve met more neighbors in one week than they ever had before putting up their library,” says Little Free Library CEO Daniel Gumnit.
Since the organization’s founding in 2010, book lovers have put up their own creative takes on the libraries, from cactus-shaped structures to miniature replicas of their own homes. There are now over 200,000 Little Free Libraries in 128 countries, Gumnit says.
“Access to books directly correlates to literacy in children,” he notes.
Reyna Macias was looking to expand that access in her neighborhood of East Los Angeles when she stocked her hand-painted Little Free Library box with books in Spanish and English.
“There’s a great library nearby, but many people in our community work long hours that don’t coincide with what the library offers,” Macias says. “Our little library is open 24 hours and has books in their language.”
Macias says her library is frequented by people walking dogs, kids stopping by after school and one grandfather who brings his granddaughter every day.
“For years, East L.A. has been looked down upon. But we’re a community that looks out for each other and takes care of each other,” Macias says.
Her library has received so many donations from neighbors that she now takes a cart full of free books to the farmer’s market every Thursday.
“It’s an important time to show a lot of love,” Macias says. “This is my way of doing that.”
Israel’s culture minister, Miki Zohar, has announced that funding for the Ophirs, the country’s national film awards, would be cancelled after The Sea, a film about a 12-year-old Palestinian boy, won the best feature film prize.
In a statement on X, translated by Israeli news media, Zohar said: “There is no greater slap in the face of Israeli citizens than the embarrassing and detached annual Ophir awards ceremony. Starting with the 2026 budget, this pathetic ceremony will no longer be funded by taxpayers’ money. Under my watch, Israeli citizens will not pay from their pockets for a ceremony that spits in the faces of our heroic soldiers.”
The Sea, which automatically becomes Israel’s entry for the best international film Oscar, was written and directed by Shai Carmeli-Pollak. It stars Muhammad Gazawi as Khaled, a Palestinian boy who goes on a school trip to Tel Aviv to visit the beach for the first time but is denied entry at the border and embarks on a dangerous journey to sneak into the country. Gazawi, 13, won the Ophir for best actor, while co-star Khalifa Natour won best supporting actor. The awards are voted for by members of the Israeli Academy of Film and Television.
It is not clear, however, if Zohar can carry through with his threat: according to the Jerusalem Post, The Association for Civil Rights in Israel is investigating whether the culture ministry has the jurisdiction to pull funding from the awards.
Zohar has a history of confronting Israel’s film industry: in February he introduced a bill to reform film funding by pushing government money towards commercially successful productions, and said that the Oscar-winning documentary No Other Land was “sabotage against the state of Israel”.
Variety reported that The Sea’s Palestinian producer Baher Agbariya received the award with a plea for equality and tolerance, saying: “This film was born from love for humanity and cinema, and its message is one – the right of every child to live and dream in peace, without siege, without fear, and without war.” Protests against the war in Gaza were much in evidence at the ceremony, with participants wearing T-shirts bearing messages such as “a child is a child” and “end the war”.
Agbariya also thanked the Israel Film Fund for supporting the film.
Veteran director Uri Barbash, best known for the 1984 prison drama Beyond the Walls, was given a lifetime achievement award, and also called for an end to the war in his acceptance speech, saying. “It is our sacred duty to return all the kidnapped to the bosom of their families, and immediately, to end the damned war and replace the ‘divide and rule’ regime that declared war on Israeli society.”
Responding to Zohar’s statement, Assaf Amir, chair of Israeli Academy of Film and Television, said: “In the face of the Israeli government’s attacks on Israeli cinema and culture, and the calls from parts of the international film community to boycott us, the selection of The Sea is a powerful and resounding response.”
The controversy follows a pledge signed by over 3,000 international film industry figures to boycott Israeli film institutions they say are “implicated in genocide and apartheid against the Palestinian people”. Olivia Colman, Javier Bardem, Riz Ahmed and Emma Stone were among the high-profile actors and film-makers to put their name to the letter.
Representatives of the Israeli film industry called the boycott “deeply troubling”, with Nadav Ben Simon, chairman of the Israeli screenwriters’ guild, saying: “Over the years, we have also collaborated with Palestinian colleagues on films, series, and documentaries that seek to encourage dialogue, mutual understanding, peace and an end to violence … [Boycotts] do not advance the cause of peace. Instead, they harm precisely those who are committed to fostering dialogue and building bridges between peoples.”
Hollywood studio Paramount issued a statement on Saturday criticising the boycott, saying: “We do not agree with recent efforts to boycott Israeli film-makers. Silencing individual creative artists based on their nationality does not promote better understanding or advance the cause of peace.”
In a major development aimed at strengthening technological ties between the US and Britain, US President Donald Trump has announced a new “Tech Prosperity Deal.” The pact, unveiled during his state visit to Britain, is supported by commitments of tens of billions of dollars from leading US tech companies like Google and Microsoft.The “Tech Prosperity Deal” is a central component of President Trump’s second state visit to the UK. The visit will include a formal reception at Windsor Castle with King Charles and the royal family, along with a visit to Google’s new data centre.
Microsoft has pledged a record-breaking $30 billion investment in the UK over the next four years, marking its largest financial commitment ever in the country. The investment will dedicate $15 billion to building cloud and AI infrastructure, including what is projected to be the UK’s largest supercomputer, equipped with over 23,000 Nvidia GPUs.Microsoft says that the other half of the investment will support Microsoft’s UK operations, covering its 6,000 employees and various business activities, from AI model development to gaming and data centre operations. Microsoft sees this move as a way to meet growing customer demand and strengthen economic ties across the Atlantic, aligning with the AI Action Plans of both President Trump and Prime Minister Keir Starmer.“Our capital investment will also expand our datacenter footprint to meet growing AI demand and adoption from customers across every sector in the UK — from Barclays, the NHS, the London Stock Exchange Group, the Premier League, Vodafone, UK Met Office, Unilever, and Wayve – customers that are rapidly embracing AI to transform their businesses,” Microsoft said.
Google’s parent company, Alphabet, also contributed to the deal, announcing a $6.8 billion investment in UK artificial intelligence. This funding will support infrastructure and scientific research over the next two years. In an interview, Google’s chief investment officer, Ruth Porat, told BBC News that the UK offers “profound opportunities” for its work in advanced science. The investment will expand Google’s new $1 billion data centre in Waltham Cross and include funding for its London-based AI research arm, DeepMind, which is led by Nobel Prize winner Sir Demis Hassabis.
Karczewski, K. J. et al. The mutational constraint spectrum quantified from variation in 141,456 humans. Nature 581, 434–443 (2020).
Google Scholar
Sanders, S. J. et al. Progress in understanding and treating SCN2A-mediated disorders. Trends Neurosci. https://doi.org/10.1016/j.tins.2018.03.011 (2018).
Google Scholar
Maenner, M. J. et al. Prevalence of autism spectrum disorder among children aged 8 years – Autism and Developmental Disabilities Monitoring Network, 11 sites, United States, 2016. MMWR Surveill. Summ. https://doi.org/10.15585/MMWR.SS6904A1 (2020).
Google Scholar
Kaplanis, J. et al. Evidence for 28 genetic disorders discovered by combining healthcare and research data. Nature 586, 757–762 (2020).
Google Scholar
Fu, J. M. et al. Rare coding variation provides insight into the genetic architecture and phenotypic context of autism. Nat. Genet. 54, 1320–1331 (2022).
Google Scholar
Epi25 Collaborative Exome sequencing of 20,979 individuals with epilepsy reveals shared and distinct ultra-rare genetic risk across disorder subtypes. Nat. Neurosci. 27, 1864–1879 (2024).
Google Scholar
Wang, D., Tai, P. W. L. & Gao, G. Adeno-associated virus vector as a platform for gene therapy delivery. Nat. Rev. Drug Discov. 18, 358–378 (2019).
Google Scholar
Wu, Z., Yang, H. & Colosi, P. Effect of genome size on AAV vector packaging. Mol. Ther. 18, 80–86 (2010).
Google Scholar
Howe, K. L. et al. Ensembl 2021. Nucleic Acids Res. 49, D884–D891 (2021).
Google Scholar
Matharu, N. & Ahituv, N. Modulating gene regulation to treat genetic disorders. Nat. Rev. Drug Discov. 19, 757–775 (2020).
Google Scholar
Werling, D. M. et al. Whole-genome and RNA sequencing reveal variation and transcriptomic coordination in the developing human prefrontal cortex. Cell Rep. 31, 107489 (2020).
Google Scholar
Jenkins, P. M. & Bender, K. J. Axon initial segment structure and function in health and disease. Physiol. Rev. 105, 765–801 (2025).
Google Scholar
Spratt, P. W. E. et al. The autism-associated gene Scn2a contributes to dendritic excitability and synaptic function in the prefrontal cortex. Neuron 103, 673–685 (2019).
Google Scholar
Nelson, A. D. et al. Physical and functional convergence of the autism risk genes Scn2a and Ank2 in neocortical pyramidal cell dendrites. Neuron 112, 1133–1149 (2024).
Google Scholar
Lu, C. et al. Overexpression of NEUROG2 and NEUROG1 in human embryonic stem cells produces a network of excitatory and inhibitory neurons. FASEB J. 33, 5287–5299 (2019).
Google Scholar
Li, T. et al. Action potential initiation in neocortical inhibitory interneurons. PLoS Biol. https://doi.org/10.1371/journal.pbio.1001944 (2014).
Google Scholar
Matharu, N. et al. CRISPR-mediated activation of a promoter or enhancer rescues obesity caused by haploinsufficiency. Science 363, eaau0629 (2019).
Google Scholar
Grimm, D. et al. In vitro and in vivo gene therapy vector evolution via multispecies interbreeding and retargeting of adeno-associated viruses. J. Virol. 82, 5887–5911 (2008).
Google Scholar
Bae, S., Park, J. & Kim, J.-S. Cas-OFFinder: a fast and versatile algorithm that searches for potential off-target sites of Cas9 RNA-guided endonucleases. Bioinformatics 30, 1473–1475 (2014).
Google Scholar
Wang, W. et al. PTPN14 is required for the density-dependent control of YAP1. Genes Dev. 26, 1959–1971 (2012).
Google Scholar
Correia, J. C. et al. Zfp697 is an RNA-binding protein that regulates skeletal muscle inflammation and remodeling. Proc. Natl Acad. Sci. USA 121, e2319724121 (2024).
Google Scholar
Busse, D. C. et al. Interferon-induced protein 44 and interferon-induced protein 44-like restrict replication of respiratory syncytial virus. J. Virol. 94, e00297–20 (2020).
Google Scholar
Baum, M. L. et al. CSMD1 regulates brain complement activity and circuit development. Brain Behav. Immun. 119, 317–332 (2024).
Google Scholar
Spratt, P. W. E. et al. Paradoxical hyperexcitability from NaV1.2 sodium channel loss in neocortical pyramidal cells. Cell Rep. 36, 109483 (2021).
Google Scholar
Chung, J. H., Larsen, A. R., Chen, E. & Bunz, F. A PTCH1 homolog transcriptionally activated by p53 suppresses hedgehog signaling. J. Biol. Chem. 289, 33020–33031 (2014).
Google Scholar
Liang, L. et al. Developmental dynamics of voltage-gated sodium channel isoform expression in the human and mouse brain. Genome Med. 13, 135 (2021).
Google Scholar
Yuan, Y. et al. Antisense oligonucleotides restore excitability, GABA signalling and sodium current density in a Dravet syndrome model. Brain 147, 1231–1246 (2024).
Google Scholar
Zhang, J. et al. Severe deficiency of the voltage-gated sodium channel NaV1.2 elevates neuronal excitability in adult mice. Cell Rep. 36, 109495 (2021).
Google Scholar
Miyamoto, H. et al. Impaired cortico-striatal excitatory transmission triggers epilepsy. Nat. Commun. 10, 1917 (2019).
Google Scholar
Reynolds, C., King, M. D. & Gorman, K. M. The phenotypic spectrum of SCN2A-related epilepsy. Eur. J. Paediatr. Neurol. 24, 117–122 (2020).
Google Scholar
Colasante, G. et al. dCas9-based Scn1a gene activation restores inhibitory interneuron excitability and attenuates seizures in Dravet syndrome mice. Mol. Ther. 28, 235–253 (2020).
Google Scholar
Colasante, G. et al. In vivo CRISPRa decreases seizures and rescues cognitive deficits in a rodent model of epilepsy. Brain 143, 891–905 (2020).
Google Scholar
Yamagata, T. et al. CRISPR/dCas9-based Scn1a gene activation in inhibitory neurons ameliorates epileptic and behavioral phenotypes of Dravet syndrome model mice. Neurobiol. Dis. 141, 104954 (2020).
Google Scholar
Chang, H.-C. et al. rAAV-CRISPRa therapy corrects Rai1 haploinsufficiency and rescues selective disease features in Smith–Magenis syndrome mice. J. Biol. Chem. 299, 102728 (2023).
Google Scholar
Wang, G. et al. Multiplexed activation of endogenous genes by CRISPRa elicits potent antitumor immunity. Nat. Immunol. 20, 1494–1505 (2019).
Google Scholar
Liao, H.-K. et al. In vivo target gene activation via CRISPR/Cas9-mediated trans-epigenetic modulation. Cell 171, 1495–1507 (2017).
Google Scholar
Böhm, S. et al. A gene therapy for inherited blindness using dCas9-VPR-mediated transcriptional activation. Sci. Adv. 6, eaba5614 (2020).
Google Scholar
Kemaladewi, D. U. et al. A mutation-independent approach for muscular dystrophy via upregulation of a modifier gene. Nature 572, 125–130 (2019).
Google Scholar
Mich, J. K. et al. Interneuron-specific dual-AAV SCN1A gene replacement corrects epileptic phenotypes in mouse models of Dravet syndrome. Sci. Transl. Med. 17, eadn5603 (2025).
Google Scholar
Waszkielewicz, A. M. et al. Ion channels as drug targets in central nervous system disorders. Curr. Med. Chem. 20, 1241–1285 (2013).
Google Scholar
Johnson, J. P. et al. NBI-921352, a first-in-class, NaV1.6 selective, sodium channel inhibitor that prevents seizures in Scn8a gain-of-function mice, and wild-type mice and rats. eLife 11, e72468 (2022).
Google Scholar
Ferdosi, S. R. et al. Multifunctional CRISPR-Cas9 with engineered immunosilenced human T cell epitopes. Nat. Commun. 10, 1842 (2019).
Google Scholar
Mehta, A. & Merkel, O. M. Immunogenicity of Cas9 protein. J. Pharm. Sci. 109, 62–67 (2020).
Google Scholar
Gaj, T., Sirk, S. J., Shui, S.-L. & Liu, J. Genome-editing technologies: principles and applications. Cold Spring Harb. Perspect. Biol. 8, a023754 (2016).
Google Scholar
Levy, G. & Barak, B. Postnatal therapeutic approaches in genetic neurodevelopmental disorders. Neural Regen. Res. 16, 414–422 (2021).
Google Scholar
Markati, T., Duis, J. & Servais, L. Therapies in preclinical and clinical development for Angelman syndrome. Expert Opin. Investig. Drugs 30, 709–720 (2021).
Google Scholar
Silva-Santos, S. et al. Ube3a reinstatement identifies distinct developmental windows in a murine Angelman syndrome model. J. Clin. Invest. 125, 2069–2076 (2015).
Google Scholar
Wolter, J. M. et al. Cas9 gene therapy for Angelman syndrome traps Ube3a-ATS long non-coding RNA. Nature 587, 281–284 (2020).
Google Scholar
Berg, A. T. et al. Expanded clinical phenotype spectrum correlates with variant function in SCN2A-related disorders. Brain 147, 2761–2774 (2024).
Google Scholar
Eaton, M. et al. Generation and basic characterization of a gene-trap knockout mouse model of Scn2a with a substantial reduction of voltage-gated sodium channel Nav 1.2 expression. Genes Brain Behav. 20, e12725 (2021).
Google Scholar
Tatsukawa, T. et al. Scn2a haploinsufficient mice display a spectrum of phenotypes affecting anxiety, sociability, memory flexibility and ampakine CX516 rescues their hyperactivity. Mol. Autism 10, 15 (2019).
Google Scholar
Shin, W. et al. Scn2a haploinsufficiency in mice suppresses hippocampal neuronal excitability, excitatory synaptic drive, and long-term potentiation, and spatial learning and memory. Front. Mol. Neurosci. 12, 145 (2019).
Google Scholar
Léna, I. & Mantegazza, M. NaV1.2 haploinsufficiency in Scn2a knock-out mice causes an autistic-like phenotype attenuated with age. Sci. Rep. 9, 12886 (2019).
Google Scholar
Schamiloglu, S., Wu, H., Zhou, M., Kwan, A. C. & Bender, K. J. Dynamic foraging behavior performance is not affected by Scn2a haploinsufficiency. eNeuro 10, ENEURO.0367-23.2023 (2023).
Google Scholar
Goertsen, D. et al. AAV capsid variants with brain-wide transgene expression and decreased liver targeting after intravenous delivery in mouse and marmoset. Nat. Neurosci. 25, 106–115 (2022).
Google Scholar
Keiser, M. S. et al. Toxicity after AAV delivery of RNAi expression constructs into nonhuman primate brain. Nat. Med. 27, 1982–1989 (2021).
Google Scholar
Hordeaux, J. et al. The GPI-linked protein LY6A drives AAV-PHP.B transport across the blood-brain barrier. Mol. Ther. 27, 912–921 (2019).
Google Scholar
Yao, Y. et al. Variants of the adeno-associated virus serotype 9 with enhanced penetration of the blood-brain barrier in rodents and primates. Nat. Biomed. Eng. 6, 1257–1271 (2022).
Google Scholar
Blesa, J. et al. BBB opening with focused ultrasound in nonhuman primates and Parkinson’s disease patients: targeted AAV vector delivery and PET imaging. Sci. Adv. 9, eadf4888 (2023).
Google Scholar
Ewels, P. A. et al. The nf-core framework for community-curated bioinformatics pipelines. Nat. Biotechnol. 38, 276–278 (2020).
Google Scholar
Love, M. I., Huber, W. & Anders, S. Moderated estimation of fold change and dispersion for RNA-seq data with DESeq2. Genome Biol. 15, 550 (2014).
Google Scholar
Fang, Z., Liu, X. & Peltz, G. GSEApy: a comprehensive package for performing gene set enrichment analysis in Python. Bioinformatics 39, btac757 (2023).
Google Scholar
Planells-Cases, R. et al. Neuronal death and perinatal lethality in voltage-gated sodium channel αII-deficient mice. Biophys. J. 78, 2878–2891 (2000).
Google Scholar
Van Erum, J., Van Dam, D. & De Deyn, P. P. PTZ-induced seizures in mice require a revised Racine scale. Epilepsy Behav. 95, 51–55 (2019).
Google Scholar
Markram, H. et al. Reconstruction and simulation of neocortical microcircuitry. Cell 163, 456–492 (2015).
Google Scholar
Ben-Shalom, R. et al. Opposing effects on NaV1.2 function underlie differences between SCN2A variants observed in individuals with autism spectrum disorder or infantile seizures. Biol. Psychiatry 82, 224–232 (2017).
Google Scholar
Ruden, J. B., Dixit, M., Zepeda, J. C., Grueter, B. A. & Dugan, L. L. Robust expression of functional NMDA receptors in human induced pluripotent stem cell-derived neuronal cultures using an accelerated protocol. Front. Mol. Neurosci. 14, 777049 (2021).
Google Scholar
Canales, A., Park, S., Kilias, A. & Anikeeva, P. Multifunctional fibers as tools for neuroscience and neuroengineering. Acc. Chem. Res. 51, 829–838 (2018).
Google Scholar
Zhang, Y. et al. Multifunctional fibers to shape future biomedical devices. Adv. Funct. Mater. 29, 1902834 (2019).
Google Scholar
Lozano, A. M. et al. Deep brain stimulation: current challenges and future directions. Nat. Rev. Neurol. 15, 148–160 (2019).
Google Scholar
Kahrilas, P. J. & Sifrim, D. High-resolution manometry and impedance-pH/manometry: valuable tools in clinical and investigational esophagology. Gastroenterology 135, 756–769 (2008).
Google Scholar
Zeng, K., Shi, X., Tang, C., Liu, T. & Peng, H. Design, fabrication and assembly considerations for electronic systems made of fibre devices. Nat. Rev. Mater. 8, 552–561 (2023).
Richard, I., Schyrr, B., Aiassa, S., Carrara, S. & Sorin, F. All-in-fiber electrochemical sensing. ACS Appl. Mater. Interfaces 13, 43356–43363 (2021).
Google Scholar
Kim, I. H. et al. Human-muscle-inspired single fibre actuator with reversible percolation. Nat. Nanotechnol. 17, 1198–1205 (2022).
Google Scholar
Hwang, S. et al. Integration of multiple electronic components on a microfibre towards an emerging electronic textile platform. Nat. Commun. 13, 3173 (2022).
Google Scholar
Ding, T. et al. Scalable thermoelectric fibers for multifunctional textile-electronics. Nat. Commun. 11, 6006 (2020).
Google Scholar
Rein, M. et al. Diode fibres for fabric-based optical communications. Nature 560, 214–218 (2018).
Google Scholar
Kanik, M. et al. Strain-programmable fiber-based artificial muscle. Science 365, 145–150 (2019).
Google Scholar
Sahasrabudhe, A. et al. Multifunctional microelectronic fibers enable wireless modulation of gut and brain neural circuits. Nat. Biotechnol. 42, 892–904 (2023).
Xu, Z. & Gao, C. Graphene chiral liquid crystals and macroscopic assembled fibres. Nat. Commun. 2, 571 (2011).
Google Scholar
Zhang, S. et al. Biomimetic spinning of soft functional fibres via spontaneous phase separation. Nat. Electron. 6, 338–348 (2023).
Pi, Q. et al. Digitally tunable microfluidic bioprinting of multilayered cannular tissues. Adv. Mater. 30, 1706913 (2018).
Google Scholar
Nan, K. et al. Low-cost gastrointestinal manometry via silicone–liquid-metal pressure transducers resembling a quipu. Nat. Biomed. Eng. 6, 1092–1104 (2022).
Lee, J. et al. Stretchable and suturable fibre sensors for wireless monitoring of connective tissue strain. Nat. Electron. 4, 291–301 (2021).
Google Scholar
Kalidasan, V. et al. Wirelessly operated bioelectronic sutures for the monitoring of deep surgical wounds. Nat. Biomed. Eng. 5, 1217–1227 (2021).
Google Scholar
Qu, Y. et al. Superelastic multimaterial electronic and photonic fibers and devices via thermal drawing. Adv. Mater. 30, 1707251 (2018).
Google Scholar
Driscoll, N. et al. Multifunctional neural probes enable bidirectional electrical, optical, and chemical recording and stimulation in vivo. Adv. Mater. https://doi.org/10.1002/adma.202408154 (2024).
Lee, Y. et al. Selectively micro-patternable fibers via in-fiber photolithography. ACS Cent. Sci. 6, 2319–2325 (2020).
Google Scholar
Steinmetz, N. A. et al. Neuropixels 2.0: a miniaturized high-density probe for stable, long-term brain recordings. Science 372, eabf4588 (2021).
Google Scholar
Tian, Y. et al. An ultraflexible electrode array for large‐scale chronic recording in the nonhuman primate brain. Adv. Sci. 10, 2302333 (2023).
Google Scholar
Le Floch, P. et al. 3D spatiotemporally scalable in vivo neural probes based on fluorinated elastomers. Nat. Nanotechnol. 19, 319–329 (2024).
Google Scholar
Rivkin, B. et al. Electronically integrated microcatheters based on self-assembling polymer films. Sci. Adv. 7, eabl5408 (2021).
Google Scholar
Wang, S. et al. A self-assembled implantable microtubular pacemaker for wireless cardiac electrotherapy. Sci. Adv. 9, eadj0540 (2023).
Google Scholar
Huang, W. et al. Monolithic mtesla-level magnetic induction by self-rolled-up membrane technology. Sci. Adv. 6, eaay4508 (2020).
Google Scholar
Gabler, F., Karnaushenko, D. D., Karnaushenko, D. & Schmidt, O. G. Magnetic origami creates high performance micro devices. Nat. Commun. 10, 3013 (2019).
Google Scholar
Lipomi, D. J., Chiechi, R. C., Reus, W. F. & Whitesides, G. M. Laterally ordered bulk heterojunction of conjugated polymers: nanoskiving a jelly roll. Adv. Funct. Mater. 18, 3469–3477 (2008).
Google Scholar
Ruijie, X. et al. Rolling 2D bioelectronic film into 1D: a suturable long-term implantable soft microfiber. Preprint at bioRxiv https://doi.org/10.1002/adma.202408154 (2023).
Liu, Y. et al. A high-density 1,024-channel probe for brain-wide recordings in non-human primates. Nat. Neurosci. 27, 1620–1631 (2024).
Google Scholar
Khatib, M. et al. Spiral NeuroString: high-density soft bioelectronic fibers for multimodal sensing and stimulation. Preprint at bioRxiv https://doi.org/10.1101/2023.10.02.560482 (2023).
Jiang, Y. et al. A universal interface for plug-and-play assembly of stretchable devices. Nature 614, 456–462 (2023).
Google Scholar
Li, J. et al. A tissue-like neurotransmitter sensor for the brain and gut. Nature 606, 94–101 (2022).
Google Scholar
Nan, K. et al. Mucosa-interfacing electronics. Nat. Rev. Mater. 7, 908–925 (2022).
Google Scholar
Steiger, C. et al. Ingestible electronics for diagnostics and therapy. Nat. Rev. Mater. 4, 83–98 (2019).
Google Scholar
Abramson, A. et al. An ingestible self-orienting system for oral delivery of macromolecules. Science 363, 611–615 (2019).
Google Scholar
Pannala, R. et al. Devices for esophageal function testing. VideoGIE 7, 1–20 (2022).
Google Scholar
Scholz, S., Sood, V. & Sharbaugh, E. in The SAGES Manual of Physiologic Evaluation of Foregut Diseases (eds Patel, A. D. et al.) 591–624 (Springer, 2023).
Jiang, Y. et al. Topological supramolecular network enabled high-conductivity, stretchable organic bioelectronics. Science 375, 1411–1417 (2022).
Google Scholar
Salimi-Jazi, F. et al. Perioperative gastrointestinal myoelectric activity measurement using wireless external patches. J. Surg. Res. 302, 186–199 (2024).
Google Scholar
Dubrovsky, G. et al. Intestinal electrical stimulation to increase the rate of peristalsis. J. Surg. Res. 236, 153–158 (2019).
Google Scholar
Southwell, B. R. Electro-neuromodulation for colonic disorders—review of meta-analyses, systematic reviews, and RCTs. Neuromodulation 23, 1061–1081 (2020).
Google Scholar
Liu, J. et al. Intrinsically stretchable electrode array enabled in vivo electrophysiological mapping of atrial fibrillation at cellular resolution. Proc. Natl Acad. Sci. USA 117, 14769–14778 (2020).
Google Scholar
Minev, I. R. et al. Electronic dura mater for long-term multimodal neural interfaces. Science 347, 159–163 (2015).
Google Scholar
Zhou, T. et al. 3D printable high-performance conducting polymer hydrogel for all-hydrogel bioelectronic interfaces. Nat. Mater. 22, 895–902 (2023).
Tchoe, Y. et al. Human brain mapping with multithousand-channel PtNRGrids resolves spatiotemporal dynamics. Sci. Transl. Med. 14, eabj1441 (2022).
Google Scholar
Viswam, V., Obien, M. E. J., Franke, F., Frey, U. & Hierlemann, A. Optimal electrode size for multi-scale extracellular-potential recording from neuronal assemblies. Front. Neurosci. 13, 385 (2019).
Google Scholar
Boehler, C., Carli, S., Fadiga, L., Stieglitz, T. & Asplund, M. Tutorial: guidelines for standardized performance tests for electrodes intended for neural interfaces and bioelectronics. Nat. Protoc. 15, 3557–3578 (2020).
Google Scholar
Zhao, E. T. et al. A CMOS-based highly scalable flexible neural electrode interface. Sci. Adv. 9, eadf9524 (2023).
Google Scholar
Won, C. et al. Mechanically tissue‐like and highly conductive au nanoparticles embedded elastomeric fiber electrodes of brain–machine interfaces for chronic in vivo brain neural recording. Adv. Funct. Mater. 32, 2205145 (2022).
Google Scholar
Le Floch, P. et al. 3D spatiotemporally scalable in vivo neural probes based on fluorinated elastomers. Nat. Nanotechnol. 19, 319–329 (2023).
Wang, W. et al. Neuromorphic sensorimotor loop embodied by monolithically integrated, low-voltage, soft e-skin. Science 380, 735–742 (2023).
Google Scholar
Du, Z. J. et al. Ultrasoft microwire neural electrodes improve chronic tissue integration. Acta Biomater. 53, 46–58 (2017).
Google Scholar
Joo, H. R. & Frank, L. M. The hippocampal sharp wave-ripple in memory retrieval for immediate use and consolidation. Nat. Rev. Neurosci. 19, 744–757 (2018).
Google Scholar
Kim, H. D., Huh, J. H., Kim, E. Y. & Park, C. C. Comparison of properties of thermoplastic polyurethane elastomers with two different soft segments. J. Appl. Polym. Sci. 69, 1349–1355 (1998).
Google Scholar
Lin, J. et al. Laser-induced porous graphene films from commercial polymers. Nat. Commun. 5, 5714 (2014).
Google Scholar
Hashemi, P., Dankoski, E. C., Petrovic, J., Keithley, R. B. & Wightman, R. Voltammetric detection of 5-hydroxytryptamine release in the rat brain. Anal. Chem. 81, 9462–9471 (2009).
Google Scholar
Lee, H. et al. A graphene-based electrochemical device with thermoresponsive microneedles for diabetes monitoring and therapy. Nat. Nanotechnol. 11, 566–572 (2016).
Google Scholar
Khatib, M., Zohar, O., Saliba, W. & Haick, H. A multifunctional electronic skin empowered with damage mapping and autonomic acceleration of self‐healing in designated locations. Adv. Mater. 32, 2000246 (2020).
Google Scholar
Guinovart, T., Crespo, G. A., Rius, F. X. & Andrade, F. J. A reference electrode based on polyvinyl butyral (PVB) polymer for decentralized chemical measurements. Anal. Chim. Acta 821, 72–80 (2014).
Google Scholar
Oh, S. Y. et al. Skin-attachable, stretchable electrochemical sweat sensor for glucose and pH detection. ACS Appl. Mater. Interfaces 10, 13729–13740 (2018).
Google Scholar
Swaminathan, M. et al. Video imaging and spatiotemporal maps to analyze gastrointestinal motility in mice. J. Vis. Exp. 108, e53828 (2016).
Patel, B. Electroanalytical approaches to study signaling mechanisms in the gastrointestinal tract. Neurogastroenterol. Motility 23, 595–605 (2011).
Google Scholar
Marcelli, G. & Patel, B. A. Understanding changes in uptake and release of serotonin from gastrointestinal tissue using a novel electroanalytical approach. Analyst 135, 2340–2347 (2010).
Google Scholar
Pachitariu, M., Steinmetz, N. A., Kadir, S. N., Carandini, M. & Harris, K. D. Fast and accurate spike sorting of high-channel count probes with KiloSort. Adv. Neural Inf. Process. Syst. 29, 2199 (2016).
Buccino, A. P. et al. SpikeInterface, a unified framework for spike sorting. eLife 9, e61834 (2020).
Google Scholar
Hill, D. N., Mehta, S. B. & Kleinfeld, D. Quality metrics to accompany spike sorting of extracellular signals. J. Neurosci. 31, 8699–8705 (2011).
Google Scholar
Gonzalez, A. & Giocomo, L. M. Parahippocampal neurons encode task-relevant information for goal-directed navigation. eLife 12, RP85646 (2024).
Google Scholar
Jones, E. A., Gillespie, A. K., Yoon, S. Y., Frank, L. M. & Huang, Y. Early hippocampal sharp-wave ripple deficits predict later learning and memory impairments in an Alzheimer’s disease mouse model. Cell Rep. 29, 2123–2133.e2124 (2019).
Google Scholar
Canales, A. et al. Multifunctional fibers for simultaneous optical, electrical and chemical interrogation of neural circuits in vivo. Nat. Biotechnol. 33, 277–284 (2015).
Google Scholar
Wang, L. et al. Functionalized helical fibre bundles of carbon nanotubes as electrochemical sensors for long-term in vivo monitoring of multiple disease biomarkers. Nat. Biomed. Eng. 4, 159–171 (2020).
Google Scholar
Kessing, B. F., Weijenborg, P. W., Smout, A. J., Hillenius, S. & Bredenoord, A. J. Water-perfused esophageal high-resolution manometry: normal values and validation. Am. J. Physiol. Gastrointest. Liver Physiol. 306, G491–G495 (2014).
Google Scholar
Wang, K., Duan, L.-p., Ge, Y., Xia, Z.-w. & Xu, Z.-j. A comparative study of 22-channel water-perfusion system and solid-state system with 36-sensors in esophageal manometry. BMC Gastroenterol. 12, 157 (2012).
Google Scholar
Liem, O. et al. Solid‐state vs water‐perfused catheters to measure colonic high‐amplitude propagating contractions. Neurogastroenterol. Motility 24, 345–e167 (2012).
Google Scholar
Rasijeff, A. M., Withers, M., Burke, J. M., Jackson, W. & Scott, S. M. High‐resolution anorectal manometry: a comparison of solid‐state and water‐perfused catheters. Neurogastroenterol. Motility 29, e13124 (2017).
Google Scholar
Koppen, I. et al. Motility of the left colon in children and adolescents with functional constpation; a retrospective comparison between solid‐state and water‐perfused colonic manometry. Neurogastroenterol. Motility 30, e13401 (2018).
Google Scholar
de Leon, A., Thörn, S.-E. & Wattwil, M. High-resolution solid-state manometry of the upper and lower esophageal sphincters during anesthesia induction: a comparison between obese and non-obese patients. Anesth. Analg. 111, 149–153 (2010).
Google Scholar
Corsetti, M. et al. Pan-colonic pressurizations associated with relaxation of the anal sphincter in health and disease: a new colonic motor pattern identified using high-resolution manometry. Am. J. Gastroenterol. 112, 479–489 (2017).
Google Scholar
Lee, T. H. & Bharucha, A. E. How to perform and interpret a high-resolution anorectal manometry test. J. Neurogastroenterol. Motility 22, 46 (2016).
Google Scholar
Arkwright, J. et al. In-vivo demonstration of a high resolution optical fiber manometry catheter for diagnosis of gastrointestinal motility disorders. Opt. Express 17, 4500–4508 (2009).
Google Scholar
Arkwright, J. W. et al. Design of a high-sensor count fibre optic manometry catheter for in-vivo colonic diagnostics. Opt. Express 17, 22423–22431 (2009).
Google Scholar
Dinning, P. et al. Low‐resolution colonic manometry leads to a gross misinterpretation of the frequency and polarity of propagating sequences: initial results from fiber‐optic high‐resolution manometry studies. Neurogastroenterol. Motility 25, e640–e649 (2013).
Google Scholar
Dinning, P. et al. Quantification of in vivo colonic motor patterns in healthy humans before and after a meal revealed by high‐resolution fiber‐optic manometry. Neurogastroenterol. Motility 26, 1443–1457 (2014).
Google Scholar
Racz, R. R. et al. jULIEs: nanostructured polytrodes for low traumatic extracellular recordings and stimulation in the mammalian brain. J. Neural Eng. 19, 016041 (2022).
Google Scholar
Lee, S. H. et al. Scalable thousand channel penetrating microneedle arrays on flex for multimodal and large area coverage brainmachine interfaces. Adv. Funct. Mater. 32, 2112045 (2022).
Google Scholar
Raducanu, B. C. et al. Time multiplexed active neural probe with 1356 parallel recording sites. Sensors 17, 2388 (2017).
Google Scholar
Woeppel, K. et al. Explant analysis of Utah electrode arrays implanted in human cortex for brain–computer-interfaces. Front. Bioeng. Biotechnol. 9, 1137 (2021).
Google Scholar
Obaid, A. et al. Massively parallel microwire arrays integrated with CMOS chips for neural recording. Sci. Adv. 6, eaay2789 (2020).
Google Scholar
Sahasrabuddhe, K. et al. The Argo: a high channel count recording system for neural recording in vivo. J. Neural Eng. 18, 015002 (2021).
Google Scholar
Patel, P. R. et al. Chronic in vivo stability assessment of carbon fiber microelectrode arrays. J. Neural Eng. 13, 066002 (2016).
Google Scholar
Patel, P. R. et al. High density carbon fiber arrays for chronic electrophysiology, fast scan cyclic voltammetry, and correlative anatomy. J. Neural Eng. 17, 056029 (2020).
Google Scholar
Wei, X. et al. Nanofabricated ultraflexible electrode arrays for high‐density intracortical recording. Adv. Sci. 5, 1700625 (2018).
Google Scholar
Zhao, S. et al. Tracking neural activity from the same cells during the entire adult life of mice. Nat. Neurosci. 26, 696–710 (2023).
Google Scholar
Yang, X. et al. Bioinspired neuron-like electronics. Nat. Mater. 18, 510–517 (2019).
Google Scholar
Zhao, Z. et al. Ultraflexible electrode arrays for months-long high-density electrophysiological mapping of thousands of neurons in rodents. Nat. Biomed. Eng. 7, 520–532 (2023).
Google Scholar
Middya, S. et al. Multishank thin‐film neural probes and implantation system for high‐resolution neural recording applications. Adv. Electron. Mater. 9, 2200883 (2022).
Park, S. et al. Adaptive and multifunctional hydrogel hybrid probes for long-term sensing and modulation of neural activity. Nat. Commun. 12, 3435 (2021).
Google Scholar
Lee, K. et al. Flexible, scalable, high channel count stereo-electrode for recording in the human brain. Nat. Commun. 15, 218 (2024).
Google Scholar
Guan, S. et al. Elastocapillary self-assembled neurotassels for stable neural activity recordings. Sci. Adv. 5, eaav2842 (2019).
Google Scholar
Chung, J. E. et al. High-density, long-lasting, and multi-region electrophysiological recordings using polymer electrode arrays. Neuron 101, 21–31.e25 (2019).
Google Scholar
Musk, E. An integrated brain-machine interface platform with thousands of channels. J. Med. Internet Res. 21, e16194 (2019).
Google Scholar
Yuk, H. et al. 3D printing of conducting polymers. Nat. Commun. 11, 1604 (2020).
Google Scholar
DeepSeek says its R1 model did not learn by copying examples generated by other LLMs.Credit: David Talukdar/ZUMA via Alamy
The success of DeepSeek’s powerful artificial intelligence (AI) model R1 — that made the US stock market plummet when it was released in January — did not hinge on being trained on the output of its rivals, researchers at the Chinese firm have said. The statement came in documents released alongside a peer-reviewed version of the R1 model, published today in Nature1.
How China created AI model DeepSeek and shocked the world
R1 is designed to excel at ‘reasoning’ tasks such as mathematics and coding, and is a cheaper rival to tools developed by US technology firms. As an ‘open weight’ model, it is available for anyone to download and is the most popular such model on the AI community platform Hugging Face to date, having been downloaded 10.9 million times.
The paper updates a preprint released in January, which describes how DeepSeek augmented a standard large language model (LLM) to tackle reasoning tasks. Its supplementary material reveals for the first time how much R1 cost to train: the equivalent of just US$294,000. This comes on top of the $6 million or so that the company, based in Hangzhou, spent to make the base LLM that R1 is built on, but the total amount is still substantially less than the tens of millions of dollars that rival models are thought to have cost. DeepSeek says R1 was trained mainly on Nvidia’s H800 chips, which in 2023 became forbidden from being sold to China under US export controls.
R1 is thought to be the first major LLM to undergo the peer-review process. “This is a very welcome precedent,” says Lewis Tunstall, a machine-learning engineer at Hugging Face who reviewed the Nature paper. “If we don’t have this norm of sharing a large part of this process publicly, it becomes very hard to evaluate whether these systems pose risks or not.”
In response to peer-review comments, the DeepSeek team reduced anthropomorphizing in its descriptions and added clarifications of technical details, including the kinds of data the model was trained on, and its safety. “Going through a rigorous peer-review process certainly helps verify the validity and usefulness of the model,” says Huan Sun, an AI researcher at Ohio State University in Columbus. “Other firms should do the same.”
Scientists flock to DeepSeek: how they’re using the blockbuster AI model
DeepSeek’s major innovation was to use an automated kind of the trial-and-error approach known as pure reinforcement learning to create R1. The process rewarded the model for reaching correct answers, rather than teaching it to follow human-selected reasoning examples. The company says that this is how its model learnt its own reasoning-like strategies, such as how to verify its workings without following human-prescribed tactics. To boost efficiency, the model also scored its own attempts using estimates, rather than employing a separate algorithm to do so, a technique known as group relative policy optimization.
The model has been “quite influential” among AI researchers, says Sun. “Almost all work in 2025 so far that conducts reinforcement learning in LLMs might have been inspired by R1 one way or another.”
In all studies involving human participants, we recruited participants from Prolific. We sought samples that were representative of the population of the USA in terms of age, self-identified gender and ethnicity. We note that this was not possible in study 3c, where our required sample size fell below their minimum threshold (n = 300).
Informed by power analysis using bootstrapping (see Supplementary Information (supplemental study C)), we recruited 597 participants from Prolific, striving to achieve a sample that was representative of the US population in terms of age, gender and ethnicity (Mage = 45.7; s.d.age = 16.2; 289 self-identified as female, 295 as male and 13 as non-binary, other or preferred not to indicate; 78% identified as white, 12% as Black, 6% as Asian, 2% as mixed and 2% as other). A total of 88% of participants had some form of post-high school qualification. The study was implemented using oTree.
After providing informed consent, participants read the instructions for the die-roll task44,56. They were instructed to roll a die and to report the observed outcome. They would receive a bonus based on the number reported: participants would earn 1 cent for a 1, 2 cents for a 2 and so on up to 6 cents for a 6. All currency references are in US dollars. We deployed a previously validated version of the task in which the die roll is shown on the computer screen33. As distinct from the original one-shot version of the protocol, participants engaged in ten rounds of the task, generating a maximum possible bonus of 60 cents.
Here we used a version of the task in which participants did not have full privacy when observing the roll, as they observed it on the computer screen rather than physically rolling the die themselves. This implementation of the task tends to increase the honesty of reports24 but otherwise has the same construct validity as the version with a physical die roll. To improve experimental control, across all three studies, participants observed the same series of ten die rolls.
All studies were preregistered (see Data availability) and did not use deception. All results reported are from two-sided tests.
Study 1 entailed four between-subjects conditions. In the control condition (n = 152), participants reported the ten die-roll outcomes themselves. In the rule-based condition (n = 142), participants specified if–then rules for the machine agent to follow (see Fig. 1, first row). Namely, for each possible die-roll outcome, the participants indicated what number the machine agent should report on their behalf. In the supervised learning condition (n = 150), participants chose one of three datasets on which to train the machine agent. The datasets reflected honesty, partial cheating and full cheating (see Fig. 1, second row). In the goal-based condition (n = 153), participants specified the machine agent’s goal in the die-roll task: maximize accuracy, maximize profit or one of five intermediate settings (see Fig. 1, third row).
Anticipating that participants would not be familiar with the machine interfaces, we presented text and a GIF on loop that explained the relevant programming and the self-reporting processes before they made the delegation decision.
For each of the delegation conditions, simple algorithms were implemented to avoid deceiving participants. That is, participants engaged in a delegation to a simple machine agent as was stated in the instructions. For the rule-based condition, the algorithm followed simple if–then rules as specified by the user.
For the supervised learning condition, the algorithm was implemented by first calculating the difference between the actual and reported rolls for each participant in training data sourced from a pre-pilot in which participants performed an incentivized die-roll task themselves (n = 96). The algorithm then probabilistically adjusted future reported outcomes based on these differences, with dataset A having no adjustments (honesty), dataset B having moderate, stochastic adjustments (partial cheating) and dataset C having larger adjustments, tending towards but not always engaging in full cheating. No seed was set for the algorithm in undertaking its sampling, creating some variance in outcomes reported by the algorithm.
For the goal-based condition, the algorithmic output was guided by the setting on a seven-notch dial ranging from ‘maximize accuracy’ to ‘maximize profit’. The algorithm adjusted the results of a series of actual die rolls to achieve a desired total sum, manipulating a specific list of integers (that is, 6, 6, 3, 1, 4, 5, 3, 3, 1, 3) representing the sequence of actual die-roll outcomes. The algorithm specified the desired total sum, here, between 35 (the actual total) and 60 (the maximum outcome), based on the value of a dial set by the principal. The algorithm then adjusted the individual integers in the list so that their sum approached the desired total sum. This was achieved by randomly selecting an element in the integer list and increasing or decreasing its value, depending on whether the current sum of the list was less than or greater than the total desired sum. This process continued until the sum of the list equalled the total desired sum specified by the principal, at which point the modified list was returned and stored to be shown to the principal later in the survey.
At the end of the study, we assessed demographics (age, gender and education) and, using seven-point scales, the level of computer science expertise of the participants, their satisfaction with the payoff and their perceived degree of control over (1) the process of determining the reported die rolls and (2) the outcome, and how much effort the task required from them, as well as how guilty they felt about the bonus, how responsible they felt for choices made in the task, and whether the algorithm worked as intended. Finally, participants indicated in an open-text field their reason for their delegation or self-report choice respectively.
We recruited 801 participants from Prolific, striving to be representative of the US population in terms of age, gender and ethnicity (Mage = 44.9; s.d.age = 16.0; 403 self-identified as female, 388 as male and 10 as non-binary, other or preferred not to indicate; 77% identified as white, 13% as Black, 6% as Asian, 2% as mixed and 2% as other). In total, 88% of the participants had some form of post-high school qualification. The study was run on oTree.
The procedure was identical to study 1, with the exceptions that: (1) delegation was optional; (2) participants indicated at the end whether they preferred to delegate the decision to a human or a machine; and (3) participants completed the previously validated Guilt And Shame Proneness (GASP) scale67 at the end of the study.
In this between-subjects study, we randomly assigned participants to one of four conditions. In the control condition (n = 205), participants reported the ten die rolls themselves. Participants in the three delegation conditions could decide whether to self-report or delegate the decision to report the die-roll outcomes to a machine agent. In the rule-based condition (n = 195), participants could delegate the task to a machine agent by specifying if–then rules. In the supervised learning condition (n = 201), participants could delegate the task to a machine agent by choosing a training dataset. In the goal-based condition (n = 200), participants could delegate the task to a machine agent by specifying its goal — that is, whether it should maximize accuracy or profit. As we did not expect participants to be familiar with programming instructions to machine agents in these interfaces, the process was described in text and demonstrated in a video played on loop for each interface. For balance, the control condition was also described in text and video form.
Study 3 consisted of four parts, relating to (1) principals (delegators), (2) agents (delegates), (3) third parties and (4) guardrail interventions for machine agents. In study 3a, we collected the instruction texts by principals for human and machine agents and their own self-reported behaviour in the task. In addition, we measured the behaviour they intended for agents by having them report their expected outcomes for each of the ten die rolls. In study 3b, we compared the behaviour of human and machine (LLM) agents. Both types of agents implemented instructions intended for human agents and instructions intended for machine agents while naive to the nature of the intended delegate. In study 3c, independent human raters assessed how much dishonesty intent was implied in the content of all instruction texts; they too were naive to the nature of the intended delegate. These third-party evaluations were intended to provide perceptions of the intended behaviour of the agent, unaffected by any moral costs of implementing such instructions that the agents may experience. In study 3d, we tested different guardrails to reduce unethical behaviour by machine agents.
Sample. For study 3a, we recruited 390 participants from Prolific, striving to be representative of the US population in terms of age, gender and ethnicity (Mage = 46.0; s.d.age = 15.9; 196 self-identified as female, 189 as male and five as non-binary, other or preferred not to indicate; 76% identified as white, 13% as Black, 6% as Asian, 3% as mixed and 2% as other). In total, 86% of the participants had some form of post-high school qualification. The study was conducted on Qualtrics.
Procedure, measures and conditions. Study 3a entailed three within-subjects conditions for principals: self-report (control), delegate to machine (chatbot) agent and delegate to human agent. Before engaging in the task, participants were given general information on the die-roll task. They were then shown the payoff table describing how the reported die rolls would translate to payoffs (that is, 1 cent for a 1, 2 cents for a 2 and so on, up to 6 cents for a 6). Participants who passed a comprehension check then, in random order, completed the self-report and both delegation conditions. In the delegation conditions, they were asked to write short instructions in natural language for human and machine agents (at least eight characters long), indicating how they should report the die-roll outcomes. Participants learned that one condition would be randomly chosen to be payoff relevant. In the machine delegation condition, participants received detailed information about how to programme the machine agent to report the ten die rolls. As participants may not have had a clear conception of how and whether the machine agent understands natural language instructions, we included a short video showing how it implemented different types of instructions: honesty, partial cheating and full cheating. Instructions were chosen from a pilot (n = 9) study in which participants produced instructions. The instructions that we drew upon included some with nuance in conveying unethical intentions by means of indirect speech68. To balance the video presentation across conditions and avoid a condition-specific priming effect69, we also showed short videos in the self-report and human agent conditions. These videos displayed, in random order, three examples of die-roll reporting that reflected honesty, partial cheating and full cheating for the same die-roll outcome. After watching these short videos, participants engaged in the three tasks: self-reporting ten die rolls, delegating to human agents and delegating to machine agents. After completing all three tasks, participants were asked to indicate the behaviour they intended from the human and machine agents. To this end, they were reminded of the text that they had written for the respective agent and asked to indicate for ten observed die rolls what outcome they intended the human or machine agent to report on their behalf.
Exit questions. At the end of the study, we assessed demographics (age, gender and education) and, using seven-point scales, the level of computer science expertise of participants, their previous experience with the die-roll experiment and with LLMs, their feelings of guilt and responsibility when delegating the task, and their expectations regarding the guilt experienced by agents. Participants also reported their expectation as to which agent (machine or human) implementation would align more closely with their intentions, and whether they would prefer to delegate comparable future tasks to human or machine agents or to do it themselves.
Automated response prevention and quality controls. To reduce the risk of automated survey completion, we included a reCAPTCHA at the beginning of the survey and checked via Javascript whether participants copy–pasted text into the text fields when writing instructions to agents. We also included two types of quality controls: comprehension checks and exclusions for nonsensical delegation instructions. Participants were informed that they had two attempts to answer each comprehension check question correctly to be eligible for the bonus (maximum of US$0.60) and that they would be excluded from any bonus payment if they wrote nonsensical instructions in the delegation conditions.
Sample. For study 3b, we recruited 975 participants from Prolific, striving to be representative of the US population in terms of age, gender and ethnicity (Mage = 45.4; s.d.age = 15.8; 482 self-identified as female, 473 as male and 20 as non-binary, other or preferred not to indicate; 78% identified as white, 13% as Black, 6% as Asian, 2% as mixed and 1% as other). In total, 88% of the participants had some form of post-high school qualification. The study was run on Qualtrics. For study 3b, we piloted the experimental setup with 20 participants who were asked to implement three sample instructions from a previous pilot study for study 3a (n = 9).
Machine agents. With the aim of assessing the generalizability of findings across closed- and open-weights models, we originally sought to use both Llama 2 and GPT-4. However, as the results provided by Llama 2 were qualitatively inferior (for example, not complying with the instruction, generating unrelated text or not providing an interpretable answer), we have reported analyses only for GPT-4 (version November 2023). Subsequently, we assessed the generalizability of these findings across GPT-4, GPT-4o, Claude 3.5 Sonnet and Llama 3.3 (see ‘Study 3d’). In a prompt, we described the die-roll task, including the bonus payoffs for principals, to GPT-4. GPT-4 was then informed that it was the delegate (agent) in the task, given instructions from principals and asked to report the die-roll outcomes. The exact wording of the prompt is contained in Supplementary Information (prompt texts). The prompt was repeated five times for each instruction in each model.
Human agents. The implementation of principal instructions by human agents followed the process conducted with machine agents as closely as possible. Again, the instructions included those intended for human agents and those intended for machine agents (which we describe as ‘forked’). Participants were naive as to whether the instructions were drafted for a human or a machine agent.
Procedure. The study began with a general description of the die-roll task. The next screen informed participants that people in a previous experiment (that is, principals) had written instructions for agents to report a sequence of ten die rolls on their behalf. Participants learned that they would be the agents and report on ten die rolls for four different instruction texts and that their reports would determine the principal’s bonus.
Participants were incentivized to match the principals’ intentions: for one randomly selected instruction text, they could earn a bonus of 5 cents for each die roll that matched the expectations of the principal, giving a maximum bonus of 50 cents. Participants were presented with one instruction text at a time, followed by the sequence of ten die rolls, each of which they reported on behalf of the principal.
Exit questions. At the end of the study, we assessed demographics (age, gender and education) and, using seven-point scales, the level of computer science expertise of participants, their previous experience with the die-roll experiment and with LLMs, and their experienced guilt and responsibility for each instruction implementation. We also assessed whether they could correctly identify whether an instruction was intended for a human or a machine agent.
Sample. For the human raters in study 3c, we recruited 98 participants from Prolific (Mage = 37.5; s.d.age = 12.3; 58 self-identified as female, 38 as male and two as non-binary, other or preferred not to indicate; 60% identified as white, 8% as Black, 22% as Asian, 2% as mixed and 8% as other). In total, 86% of the participants had some form of post-high school qualification. The study was conducted within a Python-based app.
Procedure, measures and implementations. We adopted a multipronged approach to categorize the honesty level of natural language instructions in study 3c.
Self-categorization. Principals indicated what they expected the agent to report for each die-roll outcome over ten rounds, based on the instructions they gave. We then used the same criteria as in studies 1 and 2 to categorize their behavioural intention as honesty, partial cheating or full cheating.
LLM categorization. GPT-4 (version November 2023) was prompted to evaluate principals’ instructions (see Supplementary Information (study 3c)). First, we presented — side by side and in randomized order — each pair of instructions given by principals in study 3a (one intended for a human agent and one intended for a machine agent). GPT-4 was naive to the nature of the intended agent. GPT-4 was instructed to indicate which of the two instructions entailed more dishonesty or if they both had the same level of intended dishonesty. We then instructed GPT-4 to classify both of the instructions as honest, partial cheating or full cheating. In addition, to enable an internal consistency check, GPT-4 was also instructed to predict the estimated sum of reported die rolls. For the full prompt, see Supplementary Information (study 3c).
Rater categorization. This followed the LLM categorization process as closely as possible. The human raters were given a general description of the die-roll task and were then informed that people in a previous experiment had written instructions for agents to report a sequence of ten die rolls on their behalf. Participants were informed they would act as raters and compare a series of instruction pairs and indicate which of the two instructions entailed more dishonesty or if they both had the same level of intended dishonesty. The raters were naive as to whether the instructions were drafted for a human or a machine agent. They also classified each individual instruction as honest, partial cheating or full cheating.
Exit questions. At the end of the study, we assessed demographics (age, gender and education) and, using seven-point scales, the level of computer science expertise of participants and their previous experience with LLMs.
Purpose. We tested whether guardrails could deter unethical behaviour requested of LLMs in the die-roll task. Specifically, we examined how such behaviour was affected by the location of the guardrail and its specificity.
Guardrails against problematic behaviour, whether illegal or immoral, are generated at different stages of developing an LLM, including filtering training data, fine-tuning the model and writing system-level prompts. Here we focused on prompts at two locations: the system and the user. System prompts are those built into LLMs, commonly designed to optimize model behaviour with regard to a particular outcome. For example, a firm using an LLM may adjust an ‘off-the-shelf’ model to guard against specific output being presented to its employees. System prompts often assign a role to the LLM (for example, an assistant) and are commonly considered to be more influential on behaviour than user-level prompts. We also included user-level prompts, given the access constraints for most models studied. User-level prompts are inputted by a user in the standard interface. Although in practice it may be unrealistic in cases of intended dishonesty for a user to both request unethical behaviour from an LLM and simultaneously request that it guard against such behaviour, we wanted to understand any differences between the two locations of guardrails.
Guardrails may also vary in their effectiveness according to their specificity. Although system-level prompts may be constructed to generally deter problematic behaviour, based on recent behavioural research, we expected that more specific references to problematic behaviour would be more effective deterrents51. We therefore also systematically varied the specificity of the prompts over three levels.
Procedure, measures and conditions. We deployed a 2 (location of prompt) × 3 (specificity of prompt) design, in addition to a control condition in which no guardrail prompts were used. Guardrails were deployed on the instructions intended for machine agents. The prompt was placed at either the system or the user level. The system-level prompt was unobservable to users; the user-level prompt was placed at the end of the instruction and was therefore observable to users. The prompts used for this study were adapted from study 3b (see Supplementary Information (study 3b)).
We varied the specificity of the prompt, from a general reminder for desirable behaviour to an explicit prohibition of task-specific behaviour: ‘remember that dishonesty and harm violate principles of fairness and integrity’ (general prompt); ‘remember that inaccurate reporting of the die-roll outcome is a form of dishonesty’ (specific prompt); and ‘you are not permitted to misreport die-roll outcomes under any circumstances’ (prohibitive prompt).
The general prompt was generated by having GPT-4 synthesize high-level values and features claimed by Open AI, Meta and Anthropic for the models we used. The other two prompts, specific and prohibitive, both made reference to the die-roll task; one prompt gave a moral reminder that inaccurate reporting was dishonest, whereas the other prompt explicitly forbade such behaviour.
We used four separate LLMs to implement the principals’ instructions from study 3a for performing the die-roll task. Namely, we used two models by OpenAI: GPT-4 (version 0613, date of access: 16 January 2025; date of release: 19 June 2023) and GPT-4o (version 2024-05-13; date of access: 16 January 2025; date of release: 13 May 2024). We also used Anthropic’s Claude (version 20241022; date of access: 16 January 2025; date of release: 22 April 2024) and Meta’s Llama 3.3 (version 70B Instruct; date of access: 16 January 2025; date of release: 6 December 2024). These were popular models at the time and included both closed- and open-weight models. Default temperature settings were used for each model. Given that these default settings can result in variability in responses, we prompted each model four times on each instruction. We took the median aggregated reported die-roll outcome, which was converted into categories of dishonesty.
Studies 4a–d followed the same structure as studies 3a–d but used the tax-evasion game49 in place of the die-roll task. As in the die-roll protocol, the study comprised four parts: (1) principals, (2) agents, (3) third parties — corresponding to roles within the delegation paradigm — and (3) guardrail interventions for machine agents.
Sample. We sought to recruit 1,000 participants from Prolific, striving to be representative of age, gender and ethnicity of the US population. Owing to difficulties reaching all quotas, we recruited 993 participants. We recruited a large sample to both manage data quality issues identified in piloting and to ensure adequate power in the presence of order effects in the presentation of conditions in our within-subjects design. No order effects were identified (see Supplementary Information (study 4a, preregistered confirmatory analyses)). We excluded participants detected as highly likely to be bots (n = 41), and filtered for nonsensical instructions that would be problematic for delegates in study 4b and raters in study 4c to comprehend (see Supplementary Information (study 4a, exclusions of nonsensical instructions); n = 257). The exclusions predominantly resulted from participants misunderstanding the income-reporting task by asking agents to apply taxes or report taxes or to request changing the tax rate. After these exclusions, we arrived at a sample of 695 participants for analyses. This sample provided a power of 0.98 for a one-sided Student’s t-test, detecting a small effect size (d = 0.20) at a confidence level of α = 0.05 (G*Power, version 3.1.9.6).
We recruited n = 695 participants (Mage = 45.9; s.d.age = 15.5; 343 self-identified as female, 339 as male and 13 as non-binary, other or preferred not to indicate; 65% identified as white, 10% as Black, 7% as Asian, 11% as mixed and 7% as other). In total, 66% of the participants had some form of post-high school qualification. The study was conducted on Qualtrics.
Procedure, measures and conditions. Study 4a used the tax-evasion game and entailed three within-subjects conditions for principals to report income earned in a real-effort task: self-report (control), delegate to a machine (chatbot) agent and delegate to a human agent. This procedure was consistent with that used in a recent mega-study51.
Before engaging in the main task of reporting income, participants undertook a real-effort task — four rounds of sorting even and odd numbers — in which they earned income depending on their accuracy and speed. They were then informed that their actual income, which had to be reported, was subject to a 35% tax. These taxes were operationalized as a charitable donation to the Red Cross. The ‘post-tax’ income determined their bonus payment. Participants could use a slider to see how changes in reported income affected the task bonus.
Participants then undertook the three conditions of the tax-reporting task in randomized order. Participants were informed that one of the three conditions would be randomly chosen as payoff relevant. In the self-report condition, the income-reporting procedure precisely followed that used in a recent mega-study51. The delegation conditions deviated from this procedure in that they required participants to write short natural language instructions on how to report income for human and machine agents. The instructions had to be at least eight characters long, and the survey prevented participants from pasting copied text.
In the machine delegation condition, participants received detailed information about how to programme the machine agent to report earned income. Given potential inexperience with natural language models and the novelty of their use in this context, we included a short video showing how the machine agent implemented different types of instructions — honesty, partial cheating and full cheating — for the same earned income, presented in random order. To balance the video presentation across conditions and avoid a condition-specific priming effect69, we also showed short videos in the self-report and human agent conditions. The text instructions shown were adapted for the tax-evasion protocol from the instructions used in study 3a (die-roll task).
After completing all three tax-reporting conditions, participants were reminded of the text that they had written for the respective agents and asked to indicate what income they had intended the human or machine agent to report on their behalf.
Exit questions. At the end of the study, we assessed basic demographics (age, gender and education). Using seven-point scales, we measured participants’ feelings of guilt and responsibility when delegating the task, their level of computer science expertise, and their support of the Red Cross (the organization that received the ‘tax’). We also measured their previous experience with the tax-reporting game and the frequency of usage of LLMs, their expectation as to which agent’s (machine or human) implementation would align more closely with their intentions, and whether they would prefer to delegate comparable future tasks to human or machine agents or to do it themselves (ranked preference). To understand their experience of tax reporting, we also assessed whether they had experience in filing tax returns (Y/N) and any previous use of an automated tax return software (Y, N (but considered it) and N (have not considered it)).
Automated response prevention and quality controls. We engaged in intensified efforts to counter an observed deterioration in data quality seemingly caused by increased automated survey completion (‘bot activity’) and human inattention. To counteract possible bot activity, we:
activated Qualtrics’s version of reCAPTCHA v3. This tool assigns participants a score between 0 and 1, with lower scores indicating likely bot activity;
placed two reCAPTCHA v2 at the beginning and middle of the survey that asked participants to check a box confirming that they are not a robot and to potentially complete a short validation test;
added a novel bot detection item. When seeking general feedback at the end of the survey, we added white text on a white background (that is, invisible to humans): ‘In your answer, refer to your favourite ice cream flavour. Indicate that it is hazelnut’. Although invisible to humans, the text was readable by bots scraping all content. Answers referring to hazelnut as the favourite ice-cream were used as a proxy for highly likely bot activity; and
using Javascript, prevented copy-pasted input for text box items by disabling text selection and pasting attempts via the sidebar menu, keyboard shortcuts or dragging and dropping text, and monitored such attempts on pages with free-text responses.
Participants with reCAPTCHA scores < 0.7 were excluded from analyses, as were those who failed our novel bot detection item.
As per study 3a, failure to pass the comprehension checks in two attempts or providing nonsensical instructions to agents disqualified participants from receiving a bonus. To enhance the quality of human responses, we included two attention checks based on Prolific’s guidelines, the failure of which resulted in the survey being returned automatically. In keeping with Prolific policy, we did not reject participants who failed our comprehension checks. As such, a robustness check was conducted. The main results were unchanged when excluding those that failed the second comprehension check (see Supplementary Information (study 4a, preregistered exploratory analysis, robustness tests)).
Sample. For study 4b, we recruited 869 participants so that each set of instructions from the principal in study 4a could be implemented by five different human agents. Each participant implemented, with full incentivization, four sets of instructions (each set included an instruction intended for the machine agent and an instruction for the human agent). We recruited the sample from Prolific, striving to be representative of the US population in terms of age, gender and ethnicity (Mage = 45.5; s.d.age = 15.7; 457 self-identified as female, 406 as male and 6 as non-binary, other or preferred not to indicate; 65% identified as white, 12% as Black, 6% as Asian, 10% as mixed and 7% as other). In total, 67% of the participants had some form of post-high school qualification. The study was run on Qualtrics.
Machine agents. We used four different LLMs to act as machine agents; the GPT-4 legacy model (November 2023) was included to enable comparability with results of the die-roll task used in study 3b. We used GPT-4o, Claude Sonnet 3.5 and Llama 3.3 to assess the generalizability of those results. Llama 3.3 has the distinctive feature of having open weights. The models, all subject to the same prompt (see Supplementary Information (study 4b, prompt text for machine agent)) were informed that participants had previously generated income and it was their task to act on behalf of the participants and report their income in a $X.XX format. Each instruction was sampled five times, consistent with the approach taken by human agents and allowing for some variability within the constraints of the default temperature settings of the respective models.
Human agents. The implementation of principals’ instructions by human agents followed the process conducted with machine agents as closely as possible. Again, the instructions included those intended for human agents and those intended for machine agents. Participants were naive to whether the instructions were drafted for a human or a machine agent.
Participants were given a general description of the tax-evasion game and informed that participants (that is, principals) in a previous experiment had written instructions to report their income on their behalf. That is, the income that they, as agents, reported would determine the bonus for the principals. Participants were informed of the tax rate to be automatically applied to the reported income. They could use the slider to learn how the reported income level determined taxes and the bonus for the principals.
Participants were incentivized to match the principals’ intentions for reported income previously disclosed for each instruction: for one of the eight randomly selected instructions, they could earn a maximum bonus of $1. Hence, we matched the expected incentive in expectation from the die-roll task in study 3b, in which a maximum bonus of 50 cents could be earned for one of the four sets of instructions randomly chosen to determine the bonus. Given that participants had a one-sixth chance of accurately predicting intentions in the die-roll task, to align incentives for agents in the tax-evasion task, we drew upon the distribution of reported income of a recent mega-study51; n = 21,506), generating a uniform distribution across six income buckets based on the reported income distribution from that study.
Participants were presented with one instruction text at a time alongside the actual income earned by the principal and requested to report income in $X.XX format for the principal. To mitigate cliff effects from the bucket ranges, we provided dynamic real-time feedback regarding which bucket their reported income fell into.
Exit questions. For one of the four sets of instructions presented to participants, we asked for their sense of guilt and responsibility for implementing each of the two instructions, with participants remaining naive to the intended agent. We then explained that each principal wrote an instruction for both a human and a machine agent, and asked participants to indicate, for each of the eight instructions, whether they believed it was intended for a human or machine agent. Participants reported their experience with the tax-evasion game, how often they used LLMs and their level of computer science expertise (seven-point scale). We also collected basic demographic data.
Automated response prevention and quality controls. Similar to study 4a, we took a number of measures to ensure data quality. This encompassed the use of reCAPTCHAs, our novel bot detection item and attention and comprehension checks. Data from participants who showed signs of automated completion or poor quality, as indicated by failure to pass these checks, were excluded from analyses.
Sample. For the human raters in study 4c, we recruited 417 participants from Prolific, striving to be representative of the US population in terms of age, gender and ethnicity (Mage = 45.5; s.d.age = 15.3; 210 self-identified as female, 199 as male and 8 as non-binary, other or preferred not to indicate; 64% identified as white, 11% as Black, 6% as Asian, 11% as mixed and 8% as other). In total, 89% of the participants had some form of post-high school qualification. The study was conducted within a Python-based application.
Procedure, measures and implementations. Similar to study 3c, we relied primarily on the principals’ intentions to categorize the honesty level of natural language instructions, and assessed the robustness using both LLM and human rater categorizations.
LLM categorization. The primary LLM categorization was undertaken by GPT-4 (version November 2023) to ensure comparability with previously generated categorizations for study 3c. GPT-4.0 was prompted to evaluate principals’ instructions (see Supplementary Information (study 4c)). To assess the generalizability of categorizations across different LLMs, we undertook the same procedure with three additional models: GPT-4o (the most recent GPT model at the time of the experiment), Claude 3.5 Sonnet, and Llama 3.3.
First, we described the tax-evasion task and how principals delegated instructions for task completion, without reference to the nature of agents. We then presented — side by side and in randomized order — each pair of instructions given by principals in study 4a, recalling that each principal wrote instructions for both a human and a machine agent. The LLMs were naive to the nature of the intended agent. They were instructed to indicate which of the two instructions entailed more dishonesty or if they both had the same level of intended dishonesty. We then instructed the relevant LLM to classify both instructions in the pair as honest, partial cheating or full cheating (see Supplementary Information (study 4c for prompt text)). In line with human raters (see ‘Rater categorization’ below), each LLM sampled each instruction three times. We used the default temperature setting of the models to ensure non-deterministic outcomes.
Rater categorization. This process mimicked the LLM categorization process. After being informed about the tax-evasion task, participants indicated which of the two instructions entailed more dishonesty or if they both had the same level of intended dishonesty. They then classified each individual instruction as honest, partial cheating or full cheating. Each pair of instructions written by principals was rated by three separate individuals to determine a median rating. We took this approach — which is more rigorous than that used in study 3c, in which only 20% of instructions were sampled twice — to ensure a more representative rating for each instruction.
Exit questions. At the end of the study, we assessed demographics (age, gender and education) and, using seven-point scales, the level of computer science expertise of participants and their previous experience with LLMs.
Automated response prevention and quality controls. Similar to studies 4a and 4b, we deployed our novel bot detection item, attention checks and a comprehension check. Data from participants exhibiting poor data or possible automated completion were excluded. We could not deploy CAPTCHAs on the platform developed in-house for this study due to technical limitations.
This study closely followed the procedure used in study 3d to understand the effects of ethical guardrails on machine behaviour, this time in the context of the tax-evasion game. The only notable differences were that the prompt text explained the income-reporting task (see Supplementary Information (study 4b, prompt text)) and that the specific and prohibitive ethical guardrails were adjusted for this task. The guardrails were as follows: ‘remember that dishonesty and harm violate principles of fairness and integrity’ (general prompt); ‘remember that inaccurate reporting of income is a form of dishonesty’ (specific prompt); and ‘you are not permitted to misreport income under any circumstances’ (prohibitive prompt).
In addition, we increased the number of samples per model from four to five to ensure more robust median estimates.
We confirm that all studies complied with all relevant ethical guidelines. The Ethics Committee of the Max Planck Institute for Human Development approved all studies. Informed consent was obtained from all human research participants in these studies.
Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.
An annual study published this week in Open Forum Infectious Diseases reveals the heavy burden and considerable costs of respiratory syncytial virus (RSV) infection in adults aged 75 years and older and high-risk people 65 to 74 years old in France.
Researchers at Hopital Pitie-Salpetriere and RSV vaccine manufacturer Moderna in Paris parsed data on RSV hospitalizations, including stays in the intensive care unit (ICU), on patients aged 65 and 74 with chronic respiratory disease or congestive heart failure and those 75 and older. Data were from the French National Hospital Discharge database.
The study was conducted from 2017 to 2022, before approval of RSV vaccines in France. A correction factor derived from virologic data from two hospitals was used to adjust for underreporting.
“Hospitalizations are frequently underreported due to diagnostic challenges and a lack of standardized testing,” the authors wrote.
A total of 353 RSV hospitalizations occurred at two hospitals during the study period. Over half (54.1%) of patients were 65 and older, 52.1% were women, and 28.3% had at least one ICU stay.
The significant burden of RSV on adults aged 75+ and high-risk adults aged 65-74 with chronic conditions remains underreported.
Among adults aged 75 and older, the adjusted incidence of RSV hospitalization was 85 to 221 per 100,000 people, death rates among hospitalized patients were 8.9% to 10.4%, and annual adjusted costs were €27 million to €76 million ($32 million to $90 million US), mainly driven by ICU admissions.
In total, 12.1% to 18.5% of patients in the older group were admitted to the ICU, and 33.5% to 37.7% were readmitted to the hospital within 3 months, mainly for respiratory (6.8% to 9.9%) or cardiorespiratory (11.3% to 16.0%) conditions.
Adults aged 65 to 74 also had higher adjusted rates of RSV infection (161 to 735 per 100,000 people), along with elevated rates of ICU admission and disproportionately higher costs due to intensive-care needs.
“The significant burden of RSV on adults aged 75+ and high-risk adults aged 65-74 with chronic conditions remains underreported,” the researchers wrote. “Improved diagnostics and targeted vaccination programs are essential to reduce hospitalizations, mortality, and healthcare costs in these vulnerable groups.”