Previous editions of the report relied on an expert survey. In 2026, Frontiers, a scientific publishing organization that’s co-created the Top 10 Emerging Technologies report for years, replaced it with a tool powered by large language models. Three models (GPT-4.1 mini, GPT-5, and Gemini 2.5) ran in parallel: each generated technology proposals across eight academic fields and thirteen industry sectors, cross-checking one another’s results. From more than 1,200 candidates, human experts selected eighty, and the Advisory Board chose the final ten.<\/p>
Below, we describe these ten, highlighting how AI is becoming an integral part of their development. It\u2019s no longer just a supporting technology, but a first-class component, without which the development of the solutions presented here would be hindered and, certainly, much slower.<\/p>
For people who follow AI, the biggest surprise on the list is a technology known as world models. But this isn’t just another iteration of a large language model.<\/p>
Language models learn from text: from descriptions of the world, its interpretations, its history written in words. World models learn from direct experience of the physical environment, captured by cameras, sensors and accelerometers. A video of a falling apple, a sensor signal recording the impact and a sentence describing gravity all map to the same place in a shared representation space. The model doesn’t learn the word “gravity”. It learns gravity.<\/p>
Architecturally, the key was Yann LeCun’s JEPA architecture (Joint Embedding Predictive Architecture), published in 2022. Instead of reconstructing every pixel in a video, the model learns to predict compressed representations of what will happen next: causal relationships, dynamics, the structure of events. That’s a different kind of learning than next-token prediction.<\/p>
The first major practical implementation of these ideas is the NVIDIA Cosmos platform, launched in 2025 and trained on 20 million hours of recordings from robotic, industrial and roadway environments. Robots trained on Cosmos handle physical situations they’ve never encountered before, because they reason from an internal model of matter’s behavior rather than a library of memorized cases.<\/p>
In 2026, researchers from the Stanford Doerr School of Sustainability built a world-model approach into a climate simulation, achieving a resolution that conventional models can’t reach when capturing turbulent atmospheric dynamics.<\/p>
Drug development has a structural problem: about 90% of drug candidates fail in clinical trials. Some of those failures stem from classical computers’ inability to accurately model molecular interactions. They have to simplify, and every simplification is a potential error in predicting whether a given compound will bind to the target protein. Quantum computers, by contrast, model atoms directly, using the same laws of physics that govern them.<\/p>
In 2025, IBM and Moderna completed the largest quantum simulation to date of protein folding and mRNA structure. The quantum drug discovery market doubled in value over five years. In France, startups Pasqal and Qubit Pharmaceuticals, backed by the Wellcome Trust, are advancing small-molecule discovery on neutral-atom quantum computers.<\/p>
This isn\u2019t a competition between AI and quantum computing. Quantum processors work in parallel with classical high-performance computing systems, and AI accelerates the design of biological pathways before simulations run and interprets their results. Diseases long considered too complex for classical modeling are becoming viable research targets.<\/p>
Personalized mRNA cancer vaccines are another solution that wouldn\u2019t be possible without AI.<\/p>
After a biopsy, scientists sequence the tumor’s mutations and identify proteins that distinguish cancer cells from healthy ones. AI models cut the time from a mutation profile to a finalized vaccine sequence from months to hours. The mRNA vaccine synthesized from that data trains the patient’s immune system to recognize cells carrying those markers. Each dose is unique.<\/p>
In a six-year study at Memorial Sloan Kettering Cancer Center, the survival rate for pancreatic cancer patients whose immune systems responded to the therapy was 90%. The standard five-year survival rate for this cancer is 13%. In a multicenter clinical trial in the U.S. and Europe, a personalized vaccine reduced the risk of recurrence or death in high-risk melanoma patients by 49% compared with immunotherapy alone. That combination has just entered Phase 3 clinical trials. In March 2026, the U.S. National Cancer Institute announced a $200 million public-private partnership to fund further trials.<\/p>
Exosomes are tiny membrane-bound vesicles that cells produce to shuttle proteins and genetic instructions to one another. The body recognizes them as its own, so they slip past biological barriers, including the blood-brain barrier, which most synthetic drug carriers can’t cross. Engineered exosomes can be loaded with a drug and outfitted with surface proteins that home in on a specific cell type. AI helps design these proteins and optimize the payload.<\/p>
In a clinical trial at MD Anderson Cancer Center, exosomes reached a pancreatic cancer mutation that earlier drugs hadn’t been able to target. In 2025, researchers showed that exosomes carrying gene-editing tools cross the blood-brain barrier without triggering an immune response, opening the door to therapies for Alzheimer’s, Parkinson’s and glioma. Since 2022, more than 200 clinical trials have been launched. Eli Lilly signed a $1.5 billion deal with Evox Therapeutics.<\/p>
Once quantum computers are powerful enough, they\u2019ll be able to break the algorithms protecting most of today\u2019s communications, financial transactions and medical data. In security circles, it\u2019s known as “harvest now, decrypt later”: state actors are collecting encrypted data today, banking on decrypting it in the future.<\/p>
Lattice-based cryptography addresses this threat by hiding data inside multidimensional geometric structures with deliberately introduced mathematical noise. An attacker can’t tell the correct solution from thousands of decoys. In 2024, NIST selected lattice-based algorithms as the primary standard for post-quantum encryption. ISO and the European ETSI adopted the same foundation. The EU set 2026 as the mandatory migration start date for public systems. The NSA requires post-quantum algorithms in all new national security systems starting in January 2027. Google has committed to completing its migration by 2029.<\/p>
The four remaining technologies on the list involve energy systems, materials and environmental chemistry. AI plays a supporting role in them. The breakthrough here is engineering and regulatory in nature.<\/p>
“Everything-to-grid” energy.<\/strong> Electric cars, home batteries and buildings aren\u2019t just power consumers anymore\u2014they\u2019re becoming active nodes in the grid, able to store energy and feed it back during peak demand. In 2025, Australian households installed over 180,000 home batteries, and government programs pay owners to connect them to software that coordinates collective resources. The technology exists. Tariff models and regulation are missing.<\/p> Direct lithium extraction.<\/strong> Conventional lithium production involves pumping brine into vast evaporation ponds for up to two years. New methods pull lithium from brine in a matter of hours, with recovery rates of 80\u201395% instead of the typical 50%. The first industrial facility without evaporation ponds has been operating in Argentina since 2024, at an elevation of 4,000 meters above sea level. A project in California that combines geothermal energy with lithium extraction received a $1.4 billion federal loan in February 2026.<\/p> Passive cooling materials.<\/strong> These materials radiate heat directly into outer space through the infrared atmospheric window, lowering surface temperatures below ambient without using electricity. Paints of this kind cost about $6 per square meter and deliver 15\u201320% energy savings in commercial buildings. A coating developed by the UK startup AssetCool boosts the capacity of existing power cables by 30% with no infrastructure upgrades. California and China have incorporated these materials into building codes.<\/p> PFAS destruction.<\/strong> Per- and polyfluoroalkyl substances, known as “forever chemicals”, are resistant to heat, water and chemical degradation. They’ve been detected in Arctic snow, rainwater and in the blood of nearly every person tested. Until recently, they could only be filtered, not destroyed. Several chemical and electrochemical methods have now demonstrated the ability to permanently break them down into water, carbon dioxide and mineral salts. A facility in Grand Rapids, Michigan, has been running continuously since 2023. In 2025, Japan’s Daikin destroyed more than 170,000 gallons of its own industrial wastewater. Regulatory limits aimed at protecting the environment in the EU, the U.S., Japan and Australia are turning this into a market need.<\/p> This year’s WEF list isn’t a roundup of what scientists are only now planning, but a review of what already exists in labs and early industrial deployments, now awaiting regulatory, investment and purchasing decisions.<\/p> There’s one more thing that matters. Most of the technologies on this list either wouldn\u2019t exist in their current form without AI, or wouldn\u2019t be advancing this fast. Language models cut vaccine design time from months to hours. AI-assisted quantum simulation is opening up classes of diseases that pharmacology hadn\u2019t been able to address so far, because the computational cost of modeling them was simply too high. World models let robots operate in environments no one had described to them before.<\/p> The scientific community is adopting AI fast, before most industries have even had time to ask their first strategy questions. It doesn\u2019t see it as a threat to its role, but as a tool to bring long-running, unresolved studies to a close and to pose questions no one had asked before, because their scope exceeded what existing methods could handle. This list is one of the first places where that acceleration is visible not as a declaration but as something concrete: working installations, clinical trial results, signed contracts and federal loans. AI no longer promises to change the world. It\u2019s helping change it.<\/p>AI as the world’s transformer<\/h4>