Cancer Immunotherapy Blossoms Are Starting to Bear Fruit

Over the last decade, there has been unprecedented success and progress in immunotherapy for cancer treatment, so says Ira Mellman, PhD, the president of research at the Parker Institute for Cancer Immunotherapy during his opening keynote at the Frontiers in Cancer Immunotherapy Symposium hosted by The New York Academy of Sciences, held on June 23, 2026.

Through his talk, Mellman described the research and clinical seeds that were sown over the last two decades. He showed how these seeds have blossomed, and his expectations for which blossoms will bear fruit, with the right focus and pruning.

Ira Mellman, PhD, the president of research at the Parker Institute for Cancer Immunotherapy during his opening keynote at the Frontiers in Cancer Immunotherapy Symposium hosted by The New York Academy of Sciences, held on June 23, 2026. [C. Singleman]

The last decade of cancer immunotherapy

Before considering therapeutic options, Mellman summarized the biological events required for stimulating an effective immune response in patients. He described the dual stages of T cell activation and T cell exhaustion. Most therapies have relied on interventions that activate T cells, but over time, these T cells become exhausted—they don’t die, but go into senescence.

The next logical step in therapeutic investigation was to explore methods to pull exhausted T cells from senescence back into an active state. While some new therapies are being tested, there is still significant work to be done and there have been many false starts.

Vaccine therapy has emerged to get ahead of the activation-exhaustion problem by developing primed and more efficient T cells.

“The sole purpose of vaccine therapy, as we think of it, is to generate more and better T cells by immunizing against presumptive tumor associated antigens or neo antigens and …having those cells feed into the cycle,” Mellman said. He explained that the idea behind using vaccines is akin to the goal of adaptive therapies, like CAR T, that increase the number of primed and functional T cells, but vaccines do it endogenously.

Broadening his scope further, Mellman spoke about cancers not just in terms of the direct interaction between cancer cells and T cells, but in the broader context of the body. “The tumor microenvironment (TME),” he said, “plays an immense role—both positive and negative—with respect to allowing T cells to do their jobs in the case of anti-tumor immunity.” Consideration of the interactions between cancer and the TME present both opportunities and challenges to developing therapies.

Current progress

“We’ve had unprecedented success both clinically and commercially” within the last 15 years for a variety of cancer therapies that have come from the research stage to FDA approval in a relatively short time. Mellman pointed out that these therapies not only have been created quickly, are effective across a variety of diseases, are well tolerated, and improve survival of patients.

“All of this activity together has really changed the standard of care in a wide variety of cancer types,” he stressed. “That this has happened over a 15-year period is truly extraordinary.”

Why is a renaissance needed?  

With this progress, Mellman returned to his original thesis that the field of cancer immunotherapy is on the precipice of big change.

He asked, “With this success—including clinical and financial—why do we need a renaissance in this area?” He answered his own question stating, that “We are the victims of our own success.”

He described challenges and setbacks stemming from the early progress. He described that identifying new checkpoint targets have not been successful, TIL research is somewhat successful, but outcomes are inconsistent, solid tumors post problems for therapies like CAR T, and cancer vaccines efficacy is unconfirmed except in their use as a post-operative adjuvant.

Reflecting on the history and progress, Mellman presented his four top priorities in progressing cancer immunotherapy over the next ten years. First, focusing on the next generation of cell therapy, especially focusing on treating solid tumors. Second, he suggests moving past CARs to using neoantigen-driven targeting (including vaccines, synthetic neoAg, and TCR platforms). Third, prioritizing in vivo immune engineering to reduce costs and improve accessibility and scalability. Fourth, he stresses the need to focus on a holistic approach to cancer immunology, understanding and optimizing the TME cancer interactions, developing synthetic modulators and utilizing AI models guided by patient insights.

Current steps to the future

Mellman further explained that through his work over the last two decades, he’s learned many lessons. He pointed out that researchers and clinicians need to remember that there are many therapeutic opportunities derived from various approaches; he cautioned against forgetting that mice are not humans and vice versa, pointing out that what may work in one species may not in the other; he encouraged the use of new technologies, including integrating AI into analysis procedures; and finally he suggested that researchers build on clinical research not just laboratory work.

Focusing on his own work, Mellman described how his team integrated these lessons in studying the application of mRNA vaccines. He described a study comparing the outcomes of patients with pancreatic cancer given mRNA vaccines intravenously (IV) or with an intramuscular (IM) dose. They were surprised to find that patients receiving IV delivery responded to the vaccine, but the response was inconsistent at first glance. Through deeper investigation, they determined that non-responsive patients previously had a splenectomy in addition to the removal of their pancreatic tumor. They reasoned that the IV infusion in responsive patients impacted immune cells developing in the spleen. While this study helped explain part of the process, Mellman pointed out that the fundamental mechanisms by which vaccines can be functional as an adjuvant are still unclear and his team is currently working on addressing these questions.

“There is a wealth of new understanding that we can generate if we move to the clinic quickly but do so in a way that really still concentrates on the underlying basic science,” he concluded. He stressed the importance of following approaches with an established proof of concept and suggested that engineering will enable research and scaling up to positively impact many patients.

Tending to the blossoming research and application of cancer therapy will allow for patients to have the opportunity to have better therapies in the future. Though he said this early in his discussion, it seems appropriate to end with Mellman’s assertion that, “This is a remarkable field.”

The post Cancer Immunotherapy Blossoms Are Starting to Bear Fruit appeared first on Inside Precision Medicine.

STAT+: At BIO 2026, industry wrestled with Washington politics, and making AI work better

SAN DIEGO — On the exhibition floor at the annual international BIO conference in San Diego, biotech and startup executives hummed around pavilions representing member countries and states, pausing to watch World Cup games on a giant screen at a South Korean contract drug manufacturer’s booth.

Many attendees were thinking about how to compete off the pitch, too. China’s growing power in the business of developing new drugs became a central matter for much of the convention — as well as how to boost biotech in the U.S.

The road map of hopes and fears for a U.S.-centric biotech industry followed, both onstage and off: getting a better hold on Washington and beating back pricing policies, as well as moving quickly on artificial intelligence as early strategies yield clues about how to use the technology to gain a competitive edge.

Continue to STAT+ to read the full story…

Merck KGaA to Acquire Bio-Techne for $11.3B, Expanding Life Science Tools Presence

Merck KGaA, Darmstadt, Germany, has agreed to acquire Bio-Techne for approximately $11.3 billion, the companies said today, in a deal designed to position the buyer as more of a leader across the life science value chain by expanding its presence in high-growth, next-generation life-sci markets with Bio-Techne’s tools, analytical technologies, and consumables.

The deal would add Bio-Techne’s multiomics offerings, analytical technologies, and integrated workflow solutions to German Merck’s platforms and services in research, bioprocessing and advanced therapeutics, with the aim of creating a combined company capable of helping customers from discovery and translational research through development, testing and commercial manufacturing.

Merck KGaA added that acquiring Bio-Techne would directly deliver on its mid- to long-term strategic agenda, which focuses on adding to its high-growth value drivers, integrated workflows, platformed capabilities—as well as scaling and sourcing innovation through merger-and-acquisition (M&A) deals like the Bio-Techne transaction.

That transaction is the latest in a series of acquisitions for Merck KGaA totaling more than $35 billion, including in the U.S. with acquisitions such as Millipore (for about $7 billion in 2010), as well as Sigma-Aldrich (for $17 billion in a deal announced in 2014 and completed the following year), Versum Materials (for €5.8 billion [about $6.6 billion] in 2019), and last year, SpringWorks Therapeutics (for $3.95 billion).

Merck KGaA said it would also benefit from Bio-Techne’s position as a leading provider of materials, analytics, and process technologies to cell therapy developers. Bio-Techne expects to acquire the ownership in Wilson Wolf it does not own immediately following the end of calendar year 2027 under the terms of a two-part forward contract between the company and Wilson Wolf, a manufacturer of cell culture devices, including the G-Rex product line. Bio-Techne holds 19.9% of Wilson Wolf that it acquired in the fiscal year that ended June 30, 2023.

Merck KGaA employs more than 14,000 people in the U.S. across over 70 company and customer sites.

The $11.3 billion Bio-Techne acquisition is the new third largest biopharma merger-and-acquisition (M&A) deal announced so far this year, behind the €10.7 billion ($12.268 billion) cash buyout offer for Italian-based Recordati being pursued by CVC Capital Partners and Groupe Bruxelles Lambert, which aim to take the company private; and Sun Pharmaceutical Industries’ planned $11.75 billion purchase of Organon, the women’s health drug developer spun out of Merck & Co., in a deal expected to close in early 2027.

The previous third-largest M&A deal this year, now fourth-largest, is the $10.9 billion AbbVie purchase of Apogee Therapeutics, announced on Monday. The fifth largest deal is GlaxoSmithKline (GSK)’s planned $10.6 billion buyout of Nuvalent,  announced June 9 and expected to close in the third quarter.

“Outstanding fit”

“Bio-Techne is an outstanding fit that directly supports our strategic direction focused on delivering cutting-edge products and solutions across the entire industry value chain—from lab customers to those manufacturing in the biotech and pharmaceutical industries,” Kai Beckmann, chairman of the executive board and group CEO of Merck KGaA, Darmstadt, Germany, said in a statement.

“By combining Bio-Techne’s scientific depth, innovation engine and differentiated portfolio with the global scale, manufacturing excellence and customer reach of Merck KGaA, Darmstadt, Germany, we are in a strong position to address some of the most important opportunities in life sciences and support our customers in accelerating the next generation of scientific discovery and therapeutic innovation. This positions us to deliver compelling strategic and financial benefits for shareholders, customers and employees,” Beckmann added.

Those benefits, according to German Merck, include immediate accretion to the company’s earnings before interest, taxes, depreciation, and amortization (EBITDA) pre margin for both the Group as a while and its Life Science business segment upon closing of the acquisition deal.

The Life Sciences segment finished last year with €8.98 billion ($10.36 billion) in revenue.  Merck KGaA does not break down its businesses further than its three segments, which also include healthcare (drug development, focused on oncology, neurology and immunology, and “global health” treatments such as for malaria) and electronics (high-tech materials).

The deal is expected to close by late 2026 or early 2027, subject to satisfying customary closing conditions that include obtaining regulatory approvals and approval by Bio-Techne shareholders.

Bio-Techne’s board of directors and the corporate bodies overseeing Merck KGaA, Darmstadt, Germany, have already approved the transaction, which will also add to earnings per share (EPS) by year three after closing, German Merck said.

€140M in “synergies”

Merck KGaA said it will carry out cost-cutting “synergies” of approximately €140 million (about $159.3 million) that are expected to be fully realized by the third year after closing.

The planned acquisition will be funded through a combination of existing cash on hand and proceeds from new debt, Merck KGaA said, adding that it will preserve its “strong” investment-grade credit rating.

For Minneapolis-based Bio-Techne, the acquisition is expected to increase its geographic and omnichannel access for its customers through integration of its offerings with those of Merck KGaA through a synergistic platform.

Bio-Techne has more than 3,000 employees, with approximately 2,300 employees based in the U.S. The company operates 34 global locations and 15 manufacturing facilities across the U.S., Canada, the U.K., Switzerland and China, and generated net sales of more than $1.2 billion in the fiscal year that ended June 30, 2025.

A leader in recombinant proteins with a half-century of heritage in next-generation R&D and new modalities, Bio-Techne said it would bring to German Merck a globally recognized portfolio of cytokines, growth factors, antibodies, and immunoassay kits. Bio-Techne is expected to strengthen the analytical and bioprocess solutions of Merck KGaA by adding to its offerings ProteinSimple, a leader in automated protein detection and analysis instruments. Bio-Techne added that its RNAscope and related in situ hybridization technologies would strengthen the capabilities of Merck KGaA, in spatial biology and diagnostics.

“For 50 years, Bio-Techne has enabled scientific breakthroughs across proteomics, spatial biology, and novel therapeutics,” stated Kim Kelderman, president and CEO of Bio-Techne. “This transaction is a testament to the remarkable company our team has built and to the enduring value we create for our customers and stakeholders.”

Muted enthusiasm

Bio-Techne investors appeared to share only muted enthusiasm for the deal, as the company’s shares traded on Nasdaq rose just 19.8% to $70.53 as of 12:48 pm ET, from Wednesday’s close of $58.88 per share. Merck KGaA shares traded on XETRA rose 4.93% to €147.00 ($167.25).

Puneet Souda, senior managing director, life science tools and diagnostics, and a senior research analyst with Leerink Partners, offered a possible explanation in a research note today: “The acquisition appears to be only a 24% premium to yesterday’s close and 26x the Street’s forecast for FY27 [enterprise value]/EBITDA compared to 16x for its LST [life science technologies] peer group.”

“We see the acquisition multiple undervaluing what is a highly accretive asset in our view,” Souda wrote. “Historically, TECH [Bio-Techne’s stock ticker] traded at much higher multiples given their highly accretive consumables profile (80%+ consumables) of consistent 70%+ gross margins and operating margin potential.”

One rival company in particular may benefit from the deal, Souda said: “The announcement is likely to be viewed positive for peer LST companies today, especially RVTY [Revvity] in our view.”

At $73 per share cash, the deal price represents a 36% premium to Bio-Techne’s one-month volume weighted average trading price.

“As part of Merck KGaA, Darmstadt, Germany, we will have greater scale and expanded capabilities to accelerate innovation and deepen our impact. Together, we will empower our customers to tackle the most important challenges in science and healthcare, helping to improve outcomes worldwide,” Kelderman added.

The post Merck KGaA to Acquire Bio-Techne for $11.3B, Expanding Life Science Tools Presence appeared first on GEN – Genetic Engineering and Biotechnology News.

Mental Illness Shows Context-Specific Genetic Effects

Many DNA variants linked with neuropsychiatric disorders (NPD) that do not code for proteins depend on neuronal activation, a study suggests.

The findings, in Science, highlight the power of cell stimulation to reveal context-specific “hidden” genetic effects in conditions such as schizophrenia.

They suggest that genetic regulation is not fully revealed by measuring gene expression alone.

Instead, gene activity—at least in the brain—may depend on context and the physiological state of neurons.

“Liang et al. demonstrate that the genetic processes that underlie neuropsychiatric disease are heavily determined by a dynamic physiological environment rather than by fixed cellular conditions,” said Biao Zheng, PhD, and Panos Roussos, PhD, from Icahn School of Medicine at Mount Sinai in New York, in a Perspective article accompanying the study.

They added: “To understand disease genetics, we might need to study the genome in motion and not at rest.”

Genome-wide association studies have revealed hundreds of genetic loci associated with mental illness, with more than 280 identified for schizophrenia alone.

But many of these DNA regions do not encode proteins and their impact is often subtle and difficult to detect.

To investigate further, Lifan Liang, PhD, from the University of Chicago, and co-workers studied gene expression and chromatin accessibility in single neurons derived from induced pluripotent stem cells collected from a hundred human donors.

The single-cell multi-omics study involved assessing transcriptional and epigenomic profiles before and after neurons were activated through potassium-induced depolarization.

The team found that much of the activity in regulatory DNA regions only became apparent with neuronal stimulation.

Both the number of detectable expression quantitative trait loci (eQTLs)—genetic variants associated with differences in gene expression—and chromatin accessibility QTLs (caQTLs)—DNA variants associated with differences in chromatin accessibility—rose after neuronal stimulation.

Shared and cell type-specific transcription factors worked together, possibly through regulatory cascades, to drive cell type-specific neuronal responses to stimuli.

eQTLs after stimulation had substantially weaker overlap with brain eQTL catalogs derived from postmortem tissue compared with eQTLs before stimulation.

This suggested that many relationships between regulatory DNA activity and gene expression become detectable only during neuronal activation and could be missed by traditional tissue-based studies.

A higher number of caQTLs were associated with neuropsychiatric disease compared with eQTLs, suggesting that disease-associated genetic variants could have detectable effects on regulatory DNA even when downstream changes in gene expression were not obvious.

Supporting this, chromatin accessibility and transcriptional responses to neuronal activation often occurred at different times.

Regulatory regions associated with genes that respond rapidly to neuronal stimulation often remained accessible after transcription subsided. By contrast, some late response genes exhibited accessible chromatin before their expression was induced.

When taken together, these observations implied that chromatin accessibility can be an indication of both prior and future transcriptional potential.

“We identified thousands of cell type–specific and activity-dependent quantitative trait loci for gene expression (eQTLs) and chromatin accessibility (caQTLs), helping prioritize NPD risk variants and genes that manifested functional effects only upon neuronal stimulation,” the researchers asserted.

They added: “Our work provides mechanistic insights on neuron subtype–specific activity-dependent gene regulation, substantially expanding the repertoire of context-specific causal variants and genes for NPD and other brain traits.”

The post Mental Illness Shows Context-Specific Genetic Effects appeared first on Inside Precision Medicine.

BIO 2026: FDA Leadership Confront Workforce Losses, China Competition in Drug Development

SAN DIEGO — The U.S. Food and Drug Administration (FDA) is in the middle of a cultural and operational shift that goes beyond leadership changes. U.S.–China biotechnology competition is driving discussions around regulatory reform in the U.S. where traditional paradigms are being reviewed and reconsidered, particularly for rare diseases. And patient perspectives need to be a more integral part of the drug development continuum.  

Those were some of the major themes that emerged from a town hall that took place at this year’s Biotechnology Innovation Organization (BIO) meeting in San Diego, which featured members of the current FDA leadership team. 

John Crowley, BIO president and CEO, moderated the discussion with the acting directors of Center for Drug Evaluation and Research (CDER) and the Center for Biologics Evaluation and Research (CBER) and the acting chief of staff at the FDA. During the hour-long conversation in a room packed to the hilt with BIO attendees, they spoke about the agency’s current priorities and its plans to increase its headcount, among other initiatives. 

Much of the discussion centered on ongoing plans to stabilize the agency’s workforce following the massive reduction in staffing implemented by the Department of Government Efficiency (DOGE) as well as departures of several leaders in rapid succession. The panelists acknowledged the disruptions to operations, the loss of institutional knowledge, and the past unpredictability at the agency, but did not dwell on it.  

The consensus seems to be that stabilizing the agency’s workforce is an important prerequisite for successfully launching several planned initiatives. In fact, Michael Davis, MD, PhD, acting director of CDER, noted that this has been one of his top priorities. His initial efforts were aimed at “fortifying the center and specifically the workforce” as well as finding ways to retain staff retention and boost recruitment.  

The conversation covered plans to improve overall morale, boost staff numbers, and to refocus on executing the agency’s mission. That includes implementing “some initiatives that were announced” or “have been in discussion for some time” and thinking through what is needed to support those programs, said Lowell Zeta, JD, acting chief of staff at the FDA.  

Karim Mikhail, CBER acting director, stated that in addition to working through existing submissions, his team is also planning for future challenges and ways to address them quickly to avoid backlogs.  

In terms of recruitment, the agency is looking to fill more than 2,200 authorized positions across the agency, Zeta said. About 600 people are currently being onboarded as part of the hiring push “so we feel like we’re making good progress.” CDER’s Davis said he is open to “bringing back good people” who would be interested in returning, as well as recruiting new candidates interested in public health who have the requisite skills.  

The agency is also intentional about its efforts to minimize attrition, including offering opportunities for staff to meet with leadership to discuss challenges and support needs. And those efforts may be working. In CDER, for example, staff attrition has slowed to its historical rate. 

Modernizing clinical development 

 Earlier this week, the FDA announced a slate of early actions aimed at “modernizing” and expediting early and late-stage clinical development. These were unveiled as part of Operation TrailBlazer, a U.S. Department of Health and Human Services initiative. The proposed changes are aimed at streamlining Phase I submission requirements so that drug developers have more clarity about what is necessary at this stage and what can be deferred. The agency is seeking public comments from the scientific community on some of these proposed actions.  

 The panelists positioned the proposed actions as a fundamental shift from the traditional comprehensive review approach to drug development towards a more adaptive design process. “Everybody understands the challenge we have,” Mikhail said. “We have incredible rigor” but “we need to make sure that we’re also as fast as we are rigorous.”   

Importantly, the agency is also seeking to make patient perspectives more central to the drug development process. Asked by Crowley how this will work, Davis shared an anecdote about taking part in a listening session coordinated by the FDA for parents of patients with the rare disorder, Smith-Magenis syndrome. Asking questions like “What is it like to have children with this condition? What effect does that have on the children? What effect does that have on the family dynamic?” makes it “more real when connecting the data to what families and patients are experiencing.”  

China crisis  

Another major theme here and indeed throughout the conference was maintaining U.S. competitiveness and leadership in biotech. The panelists acknowledged China’s current competitive advantage in terms of the development of its biotech infrastructure and the reality of clinical trials moving overseas due to increasing costs and the regulatory burden in the U.S.  

As Crowley put it, “China frankly is eating our lunch” and “we’re forcing so many of our innovators and companies to go to China” for early-stage clinical trials. In this climate, he noted that the FDA has a crucial role to play.  

The FDA has traditionally been viewed as the “guardian of public health, which is an important, primary role,” Crowley said, but “this notion of being a beacon of innovation and U.S. competitiveness tied to our national security is a new and important role.” The panelists also highlighted the growing use of artificial intelligence tools, digital health technologies, and wearable sensors as an important source of innovation within the agency

The FDA has recently signaled a willingness to revisit decisions it made over the past several months if those companies whose applications were rejected choose to resubmit them. “I want to make sure that we’re getting the decisions right in a way they have the confidence of the American public,” Davis said. “I think the public really trusts the FDA to make the right decisions” and “doing this closely with the multidisciplinary expert staff that we have.” 

To be clear, the agency is not going to approve everything, Mikhail said. But it will make sure that patient safety is prioritized, and that a multidisciplinary group of scientific experts at the FDA provide critical input.  

“I think everybody wants what is best for the patients,” he said. So “making sure that safety is paramount” and that “everybody is on the same page with regards to that second chance.”  

The post BIO 2026: FDA Leadership Confront Workforce Losses, China Competition in Drug Development appeared first on GEN – Genetic Engineering and Biotechnology News.

Imaging Technique Scans Entire Donor Livers for Transplant Viability

A new imaging approach to assess donor livers before transplantation could provide a much more comprehensive view of the condition of the whole organ, complementing conventional pathology tests that are typically restricted to localized areas. 

In a study published in Science Translational Medicine, researchers at the University of Oklahoma used polarization-sensitive optical coherence tomography (PS-OCT) to noninvasively measure multiple relevant parameters across the entire surface of donor livers. Using polarized light, this imaging technique was able to detect signs of steatosis, fibrosis, inflammation, and necrosis in donor livers—all of which are critical indicators of transplant viability. 

“PS-OCT offers a noninvasive assessment of liver viability by quantifying hepatic parameters across the entire donor liver, effectively complementing current pathological analysis,” write the authors of the study, led by Qinggong Tang, PhD, associate professor of biomedical engineering at the University of Oklahoma. “These results suggest that PS-OCT provides a robust approach to assessing donor liver viability, which could potentially decrease the discard rate of high-risk livers, thereby expanding the donor pool.”

Liver transplants are limited by a shortage of viable donor livers, which is driven by a high demand for donor livers and high rates of organ discard. In the U.S. alone, there are over 9,000 people on the liver transplant waitlist, which has driven healthcare providers to progressively expand the criteria used to evaluate potential donors. 

Assessing whether a donor liver is viable for transplantation currently relies on biopsies. However, these are invasive procedures that only provide information about the specific location within the liver the sample was taken from. As a result, this approach can sometimes miss critical signs of damage or disease elsewhere in the organ, potentially increasing the risk of transplanting a compromised liver. 

Tang and colleagues first evaluated the performance of PS-OCT imaging in discarded human donor livers, comparing the results to biopsies and functional tests. Using a machine learning algorithm to analyze the imaging data enabled the identification of key signs of injury and disease with 80% accuracy compared to conventional pathology assessments. The team then scanned five viable donor livers slated for transplantation and stratified them into different risk categories, accurately predicting clinical outcomes a week after transplantation. 

While biopsy tests can take several days to give back results, PS-OCT scans can produce a complete picture of the condition of the entire liver within about 15 minutes, offering a fast and noninvasive tool to assess multiple markers of transplant viability. Although the imaging technique is not intended to replace current evaluation methods, it could significantly reduce sampling error and add valuable information when screening potential donors. Ultimately, improving the assessment of donor livers could address the urgent need for increasing transplant numbers while simultaneously improving clinical outcomes for transplant recipients. 

The post Imaging Technique Scans Entire Donor Livers for Transplant Viability appeared first on Inside Precision Medicine.

Evaluating Source-Based Large Language Models for Preclinical Dermatology Education: Comparative Study

Background: Large language models (LLMs) have gained increasing popularity in medical education, with evidence supporting their educational value when framed through the lens of cognitive load theory. Source-based LLMs, which explicitly ground responses in user-uploaded material via retrieval-augmented generation algorithms, may offer additional educational value by using student-developed materials to conceptualize new areas of learning within a familiar framework. This has applications for areas like medical education in dermatology, which could benefit from inclusive sources and enhanced education to alleviate health care gaps. However, no prior studies have examined whether the inclusion of student-authored notes alters the response characteristics of a source-based LLM when responding to medical questions. Objective: This study aims to conduct an observational, comparative performance evaluation study assessing the accuracy, response reproducibility, and intermodel response similarity of freely available LLMs on text-only step 1 dermatology questions, and to explore whether providing extensive student-generated notes to a source-based LLM alters these performance characteristics. Methods: In December 2024, 4 LLMs were evaluated: NotebookLM (NLM) with uploaded preclerkship study guides (NLM w/ Notes), NLM with an uploaded blank sheet of paper (NLM w/o Notes), ChatGPT-4o Mini, and Google Gemini 1.5 Flash. Each model completed 3 trials of 121 text-based United States Medical Licensing Examination (USMLE) step 1 dermatology questions from the AMBOSS question bank. They were evaluated for overall majority-consensus accuracy, accuracy by question difficulty, intertrial reproducibility, and agreement in answer choice selection between models. Differences were analyzed through a Cochran omnibus test and subsequent pairwise McNemar tests with Benjamini-Hochberg correction. Response reproducibility and intermodel agreement were analyzed through Fleiss κ statistics with 95% CI. Results: ChatGPT-4o Mini achieved the highest overall majority-consensus accuracy (102/121, 84.3%). NLM w/ Notes demonstrated the highest intertrial reproducibility (Fleiss κ=0.927, 95% CI 0.875‐0.978) and strong performance on lower-difficulty questions but comparatively reduced accuracy on higher-difficulty items. NLM w/o Notes exhibited significantly higher omission rates (38/363, 10.5% vs ≤7/363, 1.92% for other models) than other tested LLMs. Sensitivity analysis excluding omissions increased NLM w/o Notes’ accuracy from 66.9% (81/121) to 77.8% (77/99), matching NLM w/ Notes’ accuracy of 74.4% (90/121). Intermodel agreement was significantly higher between NLM w/ Notes and ChatGPT-4o Mini compared to NLM w/o Notes and Gemini 1.5 Flash. Conclusions: Provision of student-generated notes substantially increased response reproducibility in a source-based LLM, likely reflecting consistent retrieval of similar source excerpts across trials. However, note-grounding appeared to constrain performance on higher-difficulty questions, suggesting a retrieval-augmented generation algorithm retrieval error when question stems excluded characteristic “keywords” present in lower-difficulty items. The results highlight potential challenges of a student-level, cognitive load theory–grounded educational LLM that must deal with notes not curated by experts, balance source use and internal reasoning, and meaningfully appraise uploaded sources to assess a student’s individual learning gaps.
<img src="https://jmir-production.s3.us-east-2.amazonaws.com/thumbs/bf7fd54364ca3e23ef52c35e6b9a8acd" />

The emergence of the web data infrastructure layer for AI

AI is booming. New use cases are emerging each day. To capitalize on the technology’s potential, enterprises require data at scale. In many cases, though, the relevant information is blocked or unstructured, which limits its use by AI models. 

To understand this challenge, consider the foundation of the web itself. The web was not designed for the automated discovery and retrieval that new AI applications demand. Overcoming this inherent design constraint requires infrastructure.

The next frontier in AI may depend on a new web data infrastructure layer that can enable models to discover and map this ever-expanding digital realm. This layer must be able to navigate hundreds of millions of existing web domains and billions of new URLs created each week, delivering real-time information and overcoming technical barriers.

“The data suggests there’s far more data out there,” says Or Lenchner, CEO of Bright Data, a web data collection platform. “Think of the universe: It’s out there, but you don’t know what you don’t know.”

Enabling access to fresh, relevant, and trustworthy data

While early AI breakthroughs were driven by scaling training data and model size, organizations are now encountering a fundamental bottleneck: They need to keep pace with the dynamic, unstructured, and constantly evolving nature of web data in order to ground outputs in current and verifiable information. AI performance increasingly depends not just on model architecture but on a system’s compute, networking, retrieval, and data engineering capabilities—that is, the system’s ability to quickly and reliably retrieve data that is fresh, relevant, and trustworthy.

Traditional model training relies on snapshots of information collected at a particular point in time. Training AI on such static data is no longer sufficient. To track fluctuations such as competitor pricing, consumer sentiment, and market trends, companies need a constant feed of new information, pulling data in real time along with relevant context. Their infrastructure must therefore be able to handle millions of simultaneous interactions across websites that vary by geography, language, format, and access rules.

“If it can’t retrieve real-time information, it lacks context,” Lenchner says. “In a business setting, that’s not acceptable anymore. Stale answers lead to bad decisions and disappointed consumers.”

Speed is not merely a matter of convenience; it’s a matter of necessity. Today’s organizations operate in environments where prices, inventory, markets, security threats, and customer behavior change continuously. Delayed data retrieval can reduce the usefulness of an otherwise sophisticated model.

Using live, high-quality web data can also reduce AI hallucinations because the model has a more relevant knowledge base. This builds user trust. In fact, one survey found that 56% of AI practitioners said businesses need access to real-time web data to improve trust in AI outputs. To ensure the model runs efficiently and effectively, the information must also be pared down to the appropriate essentials. 

Despite the introduction of retrieval-augmented generation (RAG), where models pull in external data at the moment of a query, many AI systems still struggle to deliver outputs that are current, contextually relevant, and trustworthy in operational settings. According to Gartner, 60% of AI projects that are not supported by AI-ready data—accurate, structured, organized, and contextualized—will be abandoned by the end of the year. 

This is because large-scale retrieval alone does not solve the problem. As Lenchner puts it, “You need to retrieve data at scale, but also in real time. Latency becomes an issue because of the end user who is waiting for the output.” 

Accessing fresh, AI-ready data at scale introduces technical and structural challenges. In practice, many enterprise systems combine public web retrieval with APIs, licensed datasets, and proprietary internal data in their AI applications. Integrating these fragmented sources into a timely and usable knowledge layer requires specialized capabilities. Some research has found that 97% of AI organizations depend on real-time web data infrastructure, but 90% feel boxed in by various restrictions. Companies are increasingly developing technical approaches to navigate these constraints.

Lenchner draws this metaphor: “Think of the trained model as intelligence and relevant data as knowledge. A powerful intelligence layer sitting on top of a hollow knowledge layer is like a genius who knows nothing—useless in practice. Intelligence and knowledge have to come together.”

The promise of new infrastructure

A new layer of web data infrastructure can address this developing need for stronger AI inputs by enabling discovery of data, real-time access, and tailoring to a specific context. As Lechner describes it, “It’s all about collecting data at scale, super-low latency, without being blocked.”

Rather than relying on increased computing power, this type of platform emulates human browsing behavior to access available content and transform raw code into structured data feeds. It can work with websites that might not interact with traditional scraping tools, such as those heavy in JavaScript, or with aggressive antibot software. 

As Lenchner explains, “It’s basically having infrastructure that can mimic a web user with identifying information—IP address, location, and 1,000 more parameters. And at scale. Think of doing that 80 billion times a day for millions of websites. And every single time, you are looking exactly as the website expects you to look.”

Of course, continuous retrieval introduces new data governance challenges. To address them, platforms can enforce strict compliance protocols aligned with global privacy frameworks, such as the EU’s General Data Protection Regulation (GDPR) and the California Consumer Privacy Act (CCPA). They can also be limited to openly accessible, public information, avoiding paywalls or private logins. Any networks used can be vetted and consent-based, and incentives can be provided to owners of IP addresses. In this way, systems can be designed to comply with tightening regulation.

Such complex capabilities do not come easy. “When this is critical infrastructure for a company,” Lenchner says, “doing it in-house becomes a full-time engineering problem that competes with the actual AI work.” Addressing this complexity requires organizations to commit significant resources, leading many to seek specialized platforms designed specifically for data retrieval, orchestration, and observability.

Infrastructure for the real world

Real-time data retrieval is changing what AI systems can do inside organizations. For example, a retail company can use public information to enable a dynamic pricing engine, and global brands can track trademark infringements. 

As the ecosystem matures, organizations that invest in this emerging data infrastructure layer will be better positioned to build AI systems that are more responsive, reliable, and aligned with real-world conditions—AI systems that can continuously adapt using current web data. Over time, the distinction between AI models and the infrastructure that feeds them may even begin to disappear.

As Lenchner says, “The world is changing. And everything that is happening in the world is being uploaded to the public web. The amount of new data that is being generated is growing and accelerating.”

To learn more from Bright Data, read the Data for AI 2026 report.

This content was produced by Insights, the custom content arm of MIT Technology Review. It was not written by MIT Technology Review’s editorial staff. It was researched, designed, and written by human writers, editors, analysts, and illustrators. This includes the writing of surveys and collection of data for surveys. AI tools that may have been used were limited to secondary production processes that passed thorough human review.