On July 15, PNAS published an international survey on how neurobiologists understand the connection between the mind and the brain, free will, and the future possibilities of their science. The final analysis included 2,657 people. The authors sent invitations to researchers who had published in neurobiology journals over the past ten years. 3,164 people responded; after checking the completeness of the questionnaires, 2,657 were included in the analysis. In the PNAS survey, 21.85% of participants partially or fully agreed that the development of neuroscience would allow for the duplication or transfer of human mental activity to a synthetic device. The question concerned the future possibilities of neuroscience. Participants were asked to assess whether knowledge of the brain could transfer mental activity to an artificial carrier. Almost every fifth participant chose to agree. In the same survey, 63.62% of participants supported reductive physicalism: they consider mental activity to be a function of the brain. Agreement with this position is encountered much more often than the expectation of uploading the mind. For the majority, physicalism has not yet turned into a forecast of transferring personality to a device. Answers about free will add another distinction. 59.26% of participants rejected the statement that the nervous system fully determines human behavior; 17.52% agreed with it. The authors associate this combination of answers with compatibilism, where free will coexists with a physical description of a person, or with the fact that the philosophical views of the participants are composed of different foundations. The survey records the views of brain researchers on several related questions: the brain creates the mind, the mind can be understood, and its transfer to a device is only allowed by a part of the participants. The authors have made the survey data and code available for further analysis of these responses.

🔗 Read original →

On July 22, a study on hematopoietic stem cells in sickle cell disease was published in Science Translational Medicine. Navitoclax restored the number of functional cells and their ability to reconstitute hematopoiesis after transplantation in mice; in patient cells outside the body, the drug improved colony formation. Gene therapy for sickle cell disease begins with a patient's own hematopoietic stem cells. They are extracted, modified with DNA or have a working gene added, and then returned after bone marrow preparation. The cells must then engraft in the bone marrow and produce new blood cells for years. The genetic construct is useless if the cell does not survive processing and engraft after transplantation. The St. Jude team studied these cells in mice and in human samples with sickle cell disease. The disease forces the bone marrow to constantly replenish damaged red blood cells. In hematopoietic stem and progenitor cells, the authors saw oxidative stress, DNA damage, and signs of senescence - a state in which a cell stops dividing and functioning normally. In a mouse model, there were fewer functional long-lived cells in the bone marrow; human cells outside the body formed colonies poorly, i.e., groups of descendants of a single cell. Then, the researchers treated the cells with navitoclax - a drug used in this study to target cells with signs of senescence. In mice, the treatment restored the number of stem cells and the ability of transplanted cells to engraft in the bone marrow. In patient samples, navitoclax or a combination of dasatinib with quercetin improved colony formation. Senescence may be impairing the very starting material for gene therapy. The authors suggest checking whether processing stem cells with senolytics improves the quality of the cell product before autologous editing, when a patient receives their own modified cells back.

🔗 Read original →

In Norway, 12 patients with early Alzheimer's underwent plasma exchange with donors aged 18-24. On July 23, a report on the open pilot was released: participants with mild cognitive impairment and Alzheimer's disease biomarkers had 16-26 liters of their own plasma replaced with plasma from 18-24 year old donors. Researchers compared three procedure regimens, their burden on patients, and adverse events. Plasma is the liquid part of blood containing proteins, hormones, and signaling molecules. During plasma exchange, an apparatus removes it from the patient and simultaneously returns the replacement fluid. In an earlier Alzheimer's study, plasma was replaced with albumin and saline; here, the replacement fluid was plasma from young donors, and the course volume reached dozens of liters. The authors built the regimen around interstitial fluid - the medium around cells in tissues. According to their model, there are about three liters of plasma in human blood flow, and about twelve in interstitial fluid. The transition of albumin and IgG between these media should take approximately 48 hours, so the intensive regimen included nine-ten procedures with three plasma packs every two-three days. Four participants underwent an average of 17.4 liters of exchange over 23.8 days. The authors call the target change in the composition of this fluid "rejuvenation of the interstitial medium". Such a course requires compatible donor plasma, an exchange apparatus, a nurse, a doctor, and monitoring of reactions. All 12 participants completed the assigned course. The study recorded 59 adverse events; among the events the authors associated with the procedure were urticaria, symptoms of calcium reduction, cough, and dizziness. Three cases of generalized urticaria stopped the current procedure and required treatment. The intensive regimen cost approximately 6,650 euros per person. The authors suggest taking it as the basis for the next controlled trial, where they will compare clinical outcomes.

🔗 Read original →

The Biomarkers of Aging Consortium will hold a colloquium on the physics of aging at Harvard on July 28. The Biomarkers of Aging Consortium has announced a one-day Gerophysics Workshop: it will take place on October 6 at Harvard Medical School in Boston. The organizers are accepting abstracts for short talks; the day will conclude with a general work session at the board. Gerophysics seeks to identify measurable quantities in aging and the rules by which they change. Observed age-related changes set state variables, and a mathematical model describes their trajectory and predicts the outcome of interventions. When two models make different predictions for one measurement, experience can distinguish between them. On the program page, the organizers ask whether biological aging should be considered an inevitable thermodynamic decay or a property of complex dynamic systems with certain constraints. The program also includes the theory of aging, limitations of living matter, energy dissipation, and irreversibility. The announced lineup includes Pyotr Fedichev, Uri Alon, Maximilian Unfried, Laszlo Barabasi, and Jesse Poganik. At the general board, participants will discuss mathematical schemes, paradoxes, and tasks for further work. This conversation already has a backstory. In a report published on May 14, 2026, participants in the first Global Conference on Gerophysics described a meeting in Singapore: on March 5-6, 2025, 160 people and 31 speakers gathered there. They proposed collecting compatible data, giving physical definitions of aging and rejuvenation, building models to predict the effects of interventions, and comparing data from different species with human research. The colloquium brings together theories of aging around a testable question: what observations and intervention outcomes should a model predict. In October, this question will be put on a separate program and discussed in a general session at the board.

🔗 Read original →

Science Corporation reported that the PRIMA system received the CE mark for sale in Europe on July 22. Science Corporation reported that DEKRA issued the CE mark to PRIMA. The company is preparing sales in 30 European countries, submitting reimbursement applications, and connecting clinics; the first commercial implant is expected in Germany. Geographic atrophy in age-related macular degeneration destroys the center of the retina - the area with which a person reads, distinguishes faces, and sees fine details. Photoreceptors there die, but the inner nerve cells of the retina are preserved in some patients. PRIMA transmits an image to them and uses them as input to the visual pathway. The surgeon places a 2x2 millimeter square implant under the retina. Glasses translate the image into near-infrared light, and the photoelements of the implant convert it into electrical impulses for the retina's neurons. The brain receives a central prosthetic image; the user changes the magnification in the glasses to decipher letters and details. In pigmentary retinitis, the MCO-010 gene therapy makes the remaining bipolar cells respond to light. PRIMA uses a different pathway: electronics deliver an external signal to the preserved retinal cells. Clinical results are described in the PRIMAvera study in the New England Journal of Medicine. In an open study without a control group with another treatment, 38 people with severe geographic atrophy received the implant; 32 underwent a year-long assessment. In 26 of them, prosthetic visual acuity improved by at least 0.2 logMAR - approximately ten letters in a standard table. The trial register notes that the result was compared to each participant's initial vision. In March, Science reported on the PRIMA application for the CE mark; now the company says the mark has been issued. Real access to the system will appear as clinics begin to perform operations, teach the use of glasses, and receive reimbursement. PRIMA adds a new input to the visual system where age-related damage has already destroyed light receptors.

🔗 Read original →

FDA Committee Recommendation

The FDA advisory committee recommended including six peptides in the list for pharmacy compounding on July 23-24. The committee advised adding BPC-157, KPV, TB-500, MOTS-c, Epitalon, and Semax to the list of substances for compounding under section 503A. Emideltide did not receive support, and the final decision will be made by the FDA.


The 503A section allows a pharmacy to prepare a medication based on a prescription for a specific patient. The 503A list defines which active substances a pharmacy can use as starting materials. Therefore, the vote concerns the possible legal path for peptides that are already being sold by longevity clinics and foreign sites as "research" substances.


Three weeks before the vote, FDA staff had recommended rejecting all seven candidates. For Epitalon, they cited a lack of publications on efficacy for insomnia, the risk of an immune reaction, and impurities in peptide synthesis. The committee took a different stance: Epitalon received 7 votes in favor and 5 against, with one abstention, while Semax received 8 votes in favor and 5 against. Emideltide was rejected with 6 votes against and 7 in favor, with one abstention. During the PCAC meeting on July 23-24, the quality of substances, clinical data, and the role of the doctor and pharmacist in individual prescriptions were discussed.


Committee member David Pope, a pharmacist at XiFin Pharmacy Solutions, explained his vote in favor of Epitalon: "The doctor and pharmacist assess the risk and decide if the peptide is suitable for a specific patient." The majority of the PCAC recommended the six substances for the 503A list. Now, the FDA will prepare a draft rule, gather public comments, and decide whether pharmacies can use these substances to manufacture medications based on individual prescriptions, as cited in FDA reports, July 2026.

🔗 Read original →

Biocomputing Ethics

On July 27, neurobiologists and bioethicists published a proposal for rules in Nature regarding laboratories that connect brain organoids to computational systems. The authors advise obtaining separate donor consent for this purpose, and if the donor cannot be reached, the project should be reviewed by an independent commission. A brain organoid is a small three-dimensional cluster of neurons grown from stem cells.


In biocomputing, researchers apply an electrical signal to the organoid through microelectrodes and read the neurons' response. The computer uses this response as a calculation result; such systems have already performed simple pattern recognition and prediction operations. Cells for organoids are often donated by people participating in medical research, and the cell line can be maintained for years and used in new projects.


According to the article's authors, consent for biomedical work does not imply consent for computational and commercial tasks, such as voice or face recognition systems. The authors suggest that laboratories create new lines with explicit consent for computational applications. For old lines, it is necessary to recontact the donor; if this is impossible, the project is evaluated by an independent commission with neurobiologists and engineers.

🔗 Read original →

Liver Support Project


Pluristyx and WFIRM are collaborating on a project to create a temporary liver support system. The team aims to establish a cell production process for a liver construct that can temporarily support a patient awaiting a transplant. The team's nearest goal is to set up the process and conduct the first test by December 2026.


The project involves measuring the production of albumin, blood clearance of bilirubin, urea synthesis, release of clotting factors, and metabolic profile to determine the functionality of the liver cells. These measurements will serve as criteria for the cell component to be accepted into the liver construct. Pluristyx is preparing a bank of induced pluripotent stem cells (iPSC) and a protocol for their conversion into liver cells.


WFIRM is developing the liver construct and testing its functionality, while ReMDO is preparing regulatory documents for commercial application. The participants have divided the tasks among themselves, with Pluristyx providing the cells, WFIRM placing them in the construct and testing its functions, and ReMDO preparing the path to clinical application, as reported in InSPA program. The project aims to create a reproducible chain for temporary liver support, with the December test showing whether the team can connect the cell bank, functional tests, and tissue construct into one process.

🔗 Read original →

Brain Data Rights

Researchers Julian Sandbrink and Michael Young published a review on implantable brain-computer interface data on July 27. They propose five guarantees: protecting de-identified records, giving individuals control over their data, separating data consent from surgical consent, limiting secondary use, and pre-defining monetization rules. An implantable brain-computer interface, or iBCI, reads brain electrical activity and translates it into commands for a cursor, speech synthesizer, prosthesis, or stimulation system.


The device collects raw signals, extracted features, decoded intentions, and algorithm parameters trained on the user's data. In their review in Communications Medicine, Sandbrink and Young describe the path of these records: from implant to clinic and manufacturer to cloud provider and algorithm developer. Treatment creates a long-term digital record of the nervous system's activity. The longer a person uses the device, the more closely it is tied to their clinical chart and decoder recognizing movement or speech.


The authors start with de-identification, proposing to protect records that may still retain individual characteristics after removing common identifiers. Patients should be given access to their data, the right to manage its transmission, and knowledge of who uses it. Consent for record storage, research, and algorithm training should be obtained separately from surgical consent. They also propose limiting secondary use that could harm individuals, such as profiling, manipulation, or discrimination. A separate issue concerns revenue from patient records: under what conditions do clinics, manufacturers, or intermediaries receive it. In the US, this topic has already led to the introduction of the MIND Act bill in the Senate.

🔗 Read original →

University Study


The University of Missouri and Altos Labs have registered a trial of sirrolimus in 225 healthy women with APOE4. On July 27, a phase I study record appeared in the ClinicalTrials.gov registry. Women aged 45–65 will be randomly assigned to either sirrolimus or a placebo; neither the participants, researchers, nor outcome assessors will know who received the treatment.


The study aims to begin enrollment in December 2026. APOE4 is a variant of the APOE gene associated with an increased risk of late-onset Alzheimer's disease. Researchers are looking for changes in the brain before memory problems appear, including measurements of brain blood flow. Sirrolimus, or rapamycin, reduces the activity of mTOR, a cellular regulator that receives signals about nutrition, growth factors, and energy storage.


The new protocol will test whether a four-week course of sirrolimus changes brain blood flow. A previous open-label pilot study by Ai-Ling Lin's group found that 23 adults without cognitive impairment who took 1 mg of sirrolimus per day for four weeks showed increased blood flow in several brain areas, with a growth of more than 15% in a subgroup of APOE4 carriers. The results will be published in Nature Aging, July 2026.

🔗 Read original →

UK Biotech Funding


The £1.6 billion funding round for Isomorphic Labs accounted for 78% of British biotech venture investments in the second quarter. According to the BioIndustry Association, British biotech companies raised £2.05 billion in venture capital from April to June, with £1.6 billion of that amount coming from Isomorphic Labs' round B funding.


The BioIndustry Association's quarterly report highlights that one deal significantly changes the market picture, with £1.6 billion out of £2.05 billion representing around 78% of all venture investments in British biotech for the second quarter of 2026. When including other equity investments, the total amount reached £2.11 billion, a five-year high.


On May 12, Isomorphic Labs announced the closure of its $2.1 billion series B round, led by Thrive Capital, with plans to use the funds to expand its work, develop its AI system for drug design, and advance programs in several therapeutic areas. The BIA reports an equivalent value of £1.6 billion for this deal, which was cited in the BioIndustry Association report.

🔗 Read original →

Ono Adopts AI Agent


Ono Pharmaceutical and Phylo announced on July 28 that researchers at the Japanese pharmaceutical company will work with Biomni Lab, an agent system for biomedicine tasks. According to the companies, it collects experiment history, works with internal data, suggests the next experience, and performs computational analysis.

The search for a drug is not just about one successful hypothesis. A researcher chooses a target, compares past experiment results with new data, decides what to test an idea with, and interprets the results. At each step, they have to search for records, launch programs, and explain to colleagues why the next experiment should be done in a certain way.

Ono Pharmaceutical and Phylo reported that Biomni Lab will enter this work with researchers searching for new drugs. A scientist formulates a task, and the agent selects a sequence of actions: links experiment history with internal data, helps plan an experiment, and launches computational biology tasks. Biomni Lab should preserve not only the answer but also the path to it, as described in an article about Biomni in Science.

Thanks to this, a researcher can see what data the analysis was based on and what led to the proposal of the next experiment. "Our researchers quickly mastered Biomni Lab and saw its potential for accelerating everyday drug discovery," said Seisi Katsumata, corporate director and executive vice president of Ono for research and drug discovery.

🔗 Read original →

Martin Jensen Warning


Martin Borch Jensen, a biotech entrepreneur, wrote on July 27 that the $2 billion invested in Altos Labs may hinder the search for a therapy through cellular reprogramming. Jensen argues that when a working method is still being sought, funds should support different attempts and allow for changing plans after experimentation.


Jensen acknowledges that Altos has significant capital, Nobel laureates, and strong leadership, but believes that after such an investment, the company is forced to present a clear plan, even though the path to therapy is still being explored through experiments. Cellular reprogramming attempts to restore a more youthful state to adult cells by temporarily activating genes that control other genes or using chemical compounds.


According to Jensen's model, a large company may choose a main hypothesis too early, hire people based on it, and build budgets and reporting around it. Then, decisions are made within the existing structure, making it more expensive to correct errors and take longer to change course. Jensen suggests giving $20 million to ten small teams, allowing them to try different approaches and increasing the stakes after achieving a working result, as described in Nature Communications and the NewLimit program.


This approach is not without strong counterarguments, including the fact that a large laboratory can also conduct multiple independent programs, as seen in the example of Bell Labs, which Jensen discusses in his response, citing single underscores such as Nature Aging, July 2026.

🔗 Read original →

Timusny peptide boosts immunotherapy


Researchers found that a thymus peptide called timulin can weaken age-related inflammation in old mice and make PD-L1 blockade more effective against their tumors. The peptide has not been administered to humans yet. The thymus is an organ where T-cells mature, and the researchers tested whether a peptide secreted by the thymus could weaken inflammation in monocytes and macrophages, cells that protect the body from infection.


As organisms age, these cells more frequently release inflammatory proteins IL-1α, IL-1β, IL-6, and TNF-α, which the authors associate with a weaker T-cell response to tumors. To isolate bone marrow cells from signals from other tissues, the researchers connected the blood flow of young and old mice and then transplanted bone marrow between them, identifying timulin as a thymus peptide whose activity decreases with age.


In a Nature Communications article from July 21, a week of timulin injections reduced the proportion of myeloid cells releasing inflammatory proteins in old mice. In cell experiments, the peptide suppressed NF-κB, a protein that activates inflammation genes. The authors also observed more inflammatory myeloid cells with age in healthy donors from 21 to 87 years old, and when they treated blood cells with timulin in a test tube, the cells released fewer of those proteins.

🔗 Read original →

Pasta Analyzes Aging


The authors from the Karolinska Institute published an article on Pasta, an open set of programs for analyzing transcriptomes, on July 27. It estimates the age shift by gene activity and suggests which chemical or genetic effects to test in cells. In the article, Jerome Sailon and colleagues trained Pasta on 17,212 samples of healthy people from 21 studies.


The transcriptome is a set of genes that a cell is using at the moment. The model compares which ones are more and less active within one sample and calculates the relative age score based on this order. One model works with different ways to measure gene activity: regular RNA sequencing, single cells, and old microchips. Usually, aging clocks give only a number: the sample looks younger or older.


But the number depends on the model: eight epigenetic clocks gave the same blood samples estimates with a spread of 17 years on average. The authors of Pasta found another application for the age score. They passed the model through the Connectivity Map, an archive of cell reactions to substances and gene changes. It contains more than three million transcriptomes of 248 cell lines after more than 30,000 chemical and 14,000 genetic effects.

🔗 Read original →

Robot Measures Cryoprotectant Thresholds


A team from Oregon State University has assembled a robot that measures the vitrification threshold of hundreds of cryoprotectant mixtures over a week. On July 22, an article was published about the setup, which consists of three 384-well plates. It determines the minimum concentration for each cryoprotectant mixture at which the solution transforms into glass, rather than ice, when cooled.


When freezing, water forms ice crystals that damage cells. Cryoprotectors prevent water molecules from gathering into crystals, making the solution glass-like. For each mixture, the concentration that suppresses ice while not making the solution too toxic for cells must be found. Typically, this search is done manually, but the authors have transferred this cycle to the three 384-well plates. The robot prepares mixtures, a camera distinguishes transparent wells with glass from those with ice, and the program selects the next concentration using a binary search.


One run yields 1,152 measurements, and after five steps, the threshold is found with an accuracy of 1% mass concentration. Over a week, the setup collected around 26,000 measurements and approximately 400 vitrification threshold values. The authors estimate its performance to be about 50 times higher than that of test tube experiments. The threshold depends not only on the solution's composition but also on the experimental conditions.

🔗 Read original →

Age Cell Analysis


Researchers trained scGPT and Geneformer on 1.3 million individual cells from four species. The models classified cells into young, middle, or old age groups with 78.6% and 80.5% accuracy. The study used cells from a fly's head, a whole worm, mouse tissues, and human blood cells.


The authors mapped genes from these species to 2,337 human analogs and tasked both models with determining the age group of each cell. scGPT reads gene activity levels, while Geneformer considers the order of genes by activity. Both models found an age signal but highlighted different gene sets.


When the data was split by donor, the accuracy ranged from 51% to 73%; when the model was trained on one species and tested on another, the accuracy dropped to 29-41%. The most consistent association was found with scGPT, which ranked RPL12, a ribosomal protein gene, as the top feature across all four species.

🔗 Read original →

Biological AI Model


Jonathan Stelzer and Amos Tanay propose building biological AI around living system processes. On July 28, Stelzer and Tanay published an article in Cell on "biological language models". The authors suggest training AI on repetitive processes of an organism and combining molecules, cells, space, and time in one model.

Many models already find patterns in DNA sequences, protein structures, and cell images, answering specific questions such as cell type, protein appearance, and risk associated with a set of features. However, experimenters often need to track the chain of events that led a cell to its state and choose an intervention in that chain.

The authors propose taking a canonical biological process - a repeating sequence of events in a living system - as the basis for models. In such a model, molecular changes, cell states, and tissue context should be combined into one sequence, rather than a set of independent measurements with a final label.

The authors call this system a "model of the world" for biology, which should link data to mechanism, showing how molecular and cell states change in space and time. The connection can be tested experimentally by changing the experimental condition and comparing the result with the model's prediction.

This idea develops the "virtual cell" program, where researchers have tried to represent and simulate molecules, cells, and tissues in different states. The new article proposes organizing these data around a process with its own dynamics, where the developer should determine which process the system explains, which measurements relate to its different moments, and what intervention will test the proposed connection.

Aging changes also pass through molecules, cells, and tissues over the years, and biological age assessment reduces the sample state to a single indicator. The process model asks a different question: what change in the cell leads to a change in tissue, and at what step can the experiment change the further course of the process.

🔗 Read original →

Biological Age Test


Ирина Конбой and Мэтт Кэберлайн discussed what a biological age test should show on July 29. Конбой suggests measuring disruptions in gene regulation, while Кэберлайн wants to check if the test result is repeatable and predicts health. Epigenetic clocks use DNA methylation and calculate age based on the pattern of these chemical marks.


Конбой proposes using the same data to measure the spread of methylation in DNA regions where the average level of methylation remains constant with age. For the same blood samples, eight epigenetic clocks differed by an average of 17 years. Конбой wants to identify specific disruptions in gene regulation in the same data, which she links to the loss of regulation.


Her approach involves measuring the spread of methylation values between people, which increases with age. This spread is what Конбой calls epigenetic noise. Her group described this approach in a 2023 article, where they selected CpG sites with consistent methylation levels and summarized the spread of values. Кэберлайн wants this signal to become a testable metric, which would require checking if the result is repeatable in parallel samples and if it predicts future organ function, disease risk, or treatment response.

🔗 Read original →

ICON Partners with Anthropic


ICON, a contract research organization, announced a multi-year partnership with Anthropic on July 28. The company plans to utilize Claude in its Orbis platform for tasks such as clinic selection, risk assessment of enrollment disruption, protocol modeling, and client access to ICON's data and expertise.


The clinical trial process often loses time before the first patient receives treatment. Sponsors select hospitals and clinics with patients who have the desired diagnosis and characteristics, then the team agrees on a protocol, including participant inclusion criteria, laboratory tests, and observation frequency. Each protocol amendment sends the document back for approval, delaying the start.


The announcement described four tasks for Claude: OneSearch and OnePlan will aid in center selection and evaluation, a predictive analytics system will identify risk signals in ongoing trials, such as enrollment deviations from the plan, and another tool will model protocol variants before document submission to identify potential future amendments. According to Pip White, Anthropic's leader in Ireland, the UK, and Northern Europe, delays are primarily due to participant enrollment: "participant enrollment is one of the biggest bottlenecks in clinical development, causing up to 80% of trials to be delayed".


The medication reaches the first patient after a series of organizational decisions, including finding a suitable center, agreeing on a protocol, and recruiting participants who meet the protocol's conditions. ICON conducts trials for pharmaceutical and biotechnology companies and plans to integrate AI into these repetitive decisions, using Claude to aid teams in making decisions before clinical data appears, specifically in center selection, enrollment planning, and protocol preparation, as reported in undisclosed clinical development publications.

🔗 Read original →

NSF PhD Pilot


The National Science Foundation of the USA has announced a pilot program, UIDP I-PhD, with a budget of $47 million over five years to support the four-year education of more than 250 STEM PhD students. The industrial partner will fund at least one year of practical work on the dissertation, and each student will have academic and industrial mentors from both the university and the company.

The NSF cites the context that among engineering, physical, computer, and information science graduates with a defined workplace, over 65% choose industry. The UIDP I-PhD program brings the university and company together at the stage when the PhD student is forming their dissertation topic and research skills. The pilot is led by the University-Industry Demonstration Partnership, an association of universities and companies.

The university funds the first year of training, while the NSF funds the subsequent years. The industrial partner takes on at least one year of practical work, where the PhD student conducts research for their dissertation on the company's premises. The company pays for the work on the dissertation and participates in scientific supervision, rather than taking the graduate student on a regular internship. Each participant will have academic and industrial mentors, and the university and company together determine how this work will be incorporated into the degree preparation.

The pilot program starts with the first cohort in fall 2026, and the NSF has allocated funds for a five-year pilot, with more than 250 PhD students participating in the first wave across the country, as reported in University-Industry Demonstration Partnership.

🔗 Read original →