AI News Archive: May 26, 2026 — Part 6
Sourced from 500+ daily AI sources, scored by relevance.
- Sam Altman thinks using AI in emails and Slack is ‘dehumanising’ – and revenue will ‘take a bit longer to figure out’
The head of one of the world’s biggest pure-play AI developers says even he’s had times when using the groundbreaking technology just wasn’t going to happen. The adoption of artificial...
- Hyper3D Launches Rodin Gen-2.5, Bringing Sculpt-Level Detail and Production Control to AI 3D Generation
Hyper3D Launches Rodin Gen-2.5, Bringing Sculpt-Level Detail and Production Control to AI 3D Generation USA Today
- Despite ‘peak hype,’ orbital data centers for AI not yet ready for NatSec prime time
Despite ‘peak hype,’ orbital data centers for AI not yet ready for NatSec prime time Breaking Defense
- Warning issued over deepfake ads that expose people to ‘the mercy of scammers’
Which? has called on the government to ensure regulator Ofcom can take action against tech firms that fail to block scams
Score: 44🌐 MovesMay 26, 2026https://www.the-independent.com/news/uk/home-news/deepfakes-social-media-scam-ai-which-ofcom-b2983728.html - Top Court Raps Pinsent Masons for 'Cavalier Attitude' to AI, Informs Regulator
The firm self-referred to the Solicitors Regulation Authority after submitting "misleading" information to the court that followed a conversation between one of the firm's lawyers and its AI tool.
- Microsoft AI coding tool costs soar: Engineers shift from Claude to GitHub Copilot
The growing financial strain highlights a broader issue facing enterprise AI adoption: the more employees use AI systems, the more expensive they become.
- 7AI launches PLAID ELITE fully managed agentic security operations service
Agentic artificial intelligence security startup 7AI Inc. today announced the launch of PLAID ELITE, a fully managed AI-native security operations service. The new service combines autonomous investigation by AI agents with oversight from 7AI security engineers to deliver continuous follow-the-sun coverage without requiring customers to build or scale an internal operations team. It handles alert ingestion, enrichment, […] The post 7AI launches PLAID ELITE fully managed agentic security operations service appeared first on SiliconANGLE .
Score: 44🌐 MovesMay 26, 2026https://siliconangle.com/2026/05/26/7ai-launches-plaid-elite-fully-managed-agentic-security-operations-service/ - As College Grads Boo Any Mention of AI, the CEO of Google Is Trying to Figure Out What to Say at an Upcoming Graduation
We suggest not mindlessly praising AI. The post As College Grads Boo Any Mention of AI, the CEO of Google Is Trying to Figure Out What to Say at an Upcoming Graduation appeared first on Futurism .
Score: 44🌐 MovesMay 26, 2026https://futurism.com/artificial-intelligence/google-ceo-speech-stanford-graduation - Rethinking organizational design in the age of agentic AI
Amid rapidly growing adoption of enterprise-level AI agents, there’s a disconnect emerging between ambition and execution. Although 85% of organizations say they want to be agentic within the next three years, 76% say their current operations and infrastructure can’t support that change. They cite a lack of readiness across people, processes, and workflows. The sticky…
Score: 44🌐 MovesMay 26, 2026https://www.technologyreview.com/2026/05/26/1137584/rethinking-organizational-design-in-the-age-of-agentic-ai/ - Workers are relying on AI more and coworkers less. It's unraveling the social fabric of work.
Workers are relying on AI more and coworkers less. It's unraveling the social fabric of work. Business Insider
Score: 44🌐 MovesMay 26, 2026https://www.businessinsider.com/ai-workplace-more-productive-less-social-2026-5 - The AI consultants coming for the Big Four
The AI consultants coming for the Big Four The Telegraph
Score: 44🌐 MovesMay 26, 2026https://www.telegraph.co.uk/business/2026/05/26/ai-consultants-coming-for-the-big-four/ - Introducing the Data 360 MCP Server — Your Unified Data, Ready for Any Agent
Turn all your Data into Context Every Agent Can Put to Work. Today, Salesforce is announcing that the Data 360 MCP Server is now available in Developer Preview — a significant step forward…
Score: 44🌐 MovesMay 26, 2026https://www.salesforce.com/blog/introducing-the-data-360-mcp-server-your-unified-data-ready-for-any-agent/ - Samsung Workers Approve Bonus Deal After Big AI Profits
Samsung Workers Approve Bonus Deal After Big AI Profits Barron's
Score: 43🌐 MovesMay 26, 2026https://www.barrons.com/news/samsung-workers-wrap-up-vote-on-massive-ai-bonus-deal-94f77f07 - New research finds all major AI models ignore faith, religion in responses
A new multi-university academic consortium led by Brigham Young University has found AI models have significant biases and gaps when it comes to addressing faith and religion. The new research from The Consortium for Evaluation of Faith and Ethics in AI (CEFE-AI)—a collaboration among researchers at BYU, Baylor University, the University of Notre Dame and Yeshiva University—found a consistent, repeatable pattern: religious perspectives are being left out of AI responses. The findings are posted to the arXiv preprint server.
Score: 43🌐 MovesMay 26, 2026https://techxplore.com/news/2026-05-major-ai-faith-religion-responses.html - Berkeley Law Implements AI Ban
“The challenge for law schools is to teach students how to effectively use AI, but to prevent it from being used to cheat on exams or papers, said Dean Erwin Chemerinsky, adding that the school is looking to “do both by incorporating it into the legal writing classes and offering courses about AI and the law.
Score: 43🌐 MovesMay 26, 2026https://feeds.feedblitz.com/~/957329927/0/law/legal-news~Berkeley-Law-Implements-AI-Ban/ - Novee debuts Agentic Fix, pushing pentest findings into Claude, Copilot and Cursor
Artificial intelligence penetration testing startup Novee Cyber Security Ltd. today launched Agentic Fix, a new capability that pushes validated exploit findings directly into the AI coding agents developers already use to write and patch software. The product extends the Novee platform by generating remediation guidance from the same exploit context used to uncover a vulnerability. […] The post Novee debuts Agentic Fix, pushing pentest findings into Claude, Copilot and Cursor appeared first on SiliconANGLE .
Score: 43🌐 MovesMay 26, 2026https://siliconangle.com/2026/05/26/novee-debuts-agentic-fix-pushing-pentest-findings-claude-copilot-cursor/ - Big Four consulting has 2 AI nightmares. KPMG’s answer to both is the same
Big Four consulting has 2 AI nightmares. KPMG’s answer to both is the same Fortune
Score: 43🌐 MovesMay 26, 2026https://fortune.com/2026/05/26/kpmg-anthropic-claude-partnership-big-four-ai/ - Less than 10% of Chinese public worried about AI destroying jobs: survey
China is better positioned than many other nations to lead in artificial intelligence due to the public’s “strikingly positive” attitude towards the technology, according to a new survey by University College London. Less than 10 per cent of respondents in China worried that AI would make it harder to find a job and about one-third believed the technology would create more high-skilled work, the survey found. A whopping 96 per cent of Chinese people surveyed said they used AI at work every week...
- Your AI Won’t Scale Without a Shared Language
Your AI Won’t Scale Without a Shared Language Boston Consulting Group
Score: 43🌐 MovesMay 26, 2026https://www.bcg.com/publications/2026/your-ai-wont-scale-without-a-shared-language - Quanscient lands €10M to advance AI- and quantum-native hardware engineering
Quanscient, a Finnish company focusedon cloud-based multiphysics simulation technology and quantum algorithms, hasraised €10 million in a Series A funding round to support its internationalexpansion a...
Score: 43💰 MoneyMay 26, 2026https://tech.eu/2026/05/26/quanscient-lands-eur10m-to-advance-ai-and-quantum-native-hardware-engineering/ - I’m a Professional Fact-Checker. AI Is Wrong More Often Than You Think
Can AI do fact-checking? A WIRED fact-checker fact-checks.
- The Real Question in AI Today: Do We Need More Giant Models?
By Atul Rai, Co-founder & CEO, Staqu Technologies Artificial intelligence has entered what appears to be a race of scale. Every few weeks, the industry sees the launch of another […] The post The Real Question in AI Today: Do We Need More Giant Models? appeared first on Express Computer .
Score: 43🌐 MovesMay 26, 2026https://www.expresscomputer.in/news/the-real-question-in-ai-today-do-we-need-more-giant-models/135435/ - S-Oil launches AI data center cooling pilot
S-Oil said Tuesday it has launched a pilot project to test liquid immersion cooling technology for AI data centers, as surging AI workloads drive up power consumption and heat generation. The pilot, designed to replicate real-world data center conditions, aims to verify the performance and operational safety of next-generation cooling systems. S-Oil will supply its immersion cooling fluid, “S-Oil E-Cooling Solution,” and provide technical support for the project. The pilot also includes particip
- Characterization of GPU-based Inference for Reasoning-Centric LLMs (Micron, Argonne)
Researchers from Micron Technology and Argonne National Laboratory have released “Understanding Inference Scaling for LLMs: Bottlenecks, Trade-offs, and Performance Principles”. Abstract “The transition from standard generative AI to reasoning-centric architectures, exemplified by models capable of extensive Chain-of-Thought (CoT) processing, marks a fundamental paradigm shift in system requirements. Unlike traditional workloads dominated by compute-bound prefill, reasoning... » read more The post Characterization of GPU-based Inference for Reasoning-Centric LLMs (Micron, Argonne) appeared first on Semiconductor Engineering .
Score: 42🌐 MovesMay 26, 2026https://semiengineering.com/characterization-of-gpu-based-inference-for-reasoning-centric-llms-micron-argonne/ - Why are big AI companies embedding engineers with customers, and what does that mean?
The promise of frontier AI has always sounded like a utility: abundant intelligence, available on demand, as easy to access as electricity, water, or cloud computing . The metaphor is powerful, and for good reason. Utilities scale because they abstract complexity away. You don’t need an engineer from the power company sitting in your office every time you turn on the lights. And yet, the most sophisticated AI companies in the world are increasingly doing something very different: They are sending people. OpenAI recently announced the OpenAI Deployment Company , explicitly designed to embed forward deployed engineers (FDE) inside organizations working on complex problems in demanding environments. These engineers, according to OpenAI , will work with business leaders, operators, and frontline teams to identify where AI can make the biggest impact, redesign workflows, and turn those gains into durable systems. Anthropic is hiring FDEs for its applied AI team , people who embed directly with strategic customers to drive enterprise adoption and ship real-world applications. And Google is doing the same . Is that a coincidence? That is revealing. Because if intelligence were already a true utility, this would not be necessary. You would not need to send your own engineers to every customer to make the faucet work. The paradox of AI as a utility This is the paradox at the heart of the current enterprise AI model: The industry speaks the language of scale, abundance, and platforms, but its delivery model increasingly resembles high-end consulting. That does not mean the work is unimportant. Quite the opposite. Forward deployed engineers are often solving the real problem: taking frontier models out of the demo environment and making them function inside messy, regulated, fragmented organizations. They deal with permissions, legacy systems, compliance, data quality, workflows, operational constraints, and all the things that make companies different from benchmarks. But that is precisely the point. The need for these people is not merely a commercial innovation. It is a symptom. It tells us that the product, as currently packaged, is not yet enough. In the previous articles in this series, I argued that large language models were never built to run a company , that enterprise AI must move from tools to systems , and that the systems that finally work will not look like chatbots or copilots, but like intelligence embedded into the organization itself . The FDE phenomenon confirms that argument from the vendor side. If the AI lab has to send engineers to reconstruct context, redesign workflows, and make the system operate under real constraints, then the missing layer is not imagined. It is sitting there, being supplied manually. The preplatform pattern Every major technology industry goes through an artisanal phase before it becomes industrial. Before enterprise software became packaged, implementation was bespoke. Before cloud platforms became mature, companies needed armies of specialists to configure infrastructure. Before the web stabilized around browsers, standards, hosting providers, content management systems, analytics, and design conventions, building a website required far more custom work than it later would. Forward deployed engineering belongs to that same historical pattern. Palantir popularized the model years ago . Its own description of the forward deployed software engineer role is based on engineers working directly inside customer environments to make software function in operational reality. That model made sense for Palantir because its customers often had extremely complex, high-stakes, highly specific environments. But when OpenAI and Anthropic begin to converge on similar patterns , the signal is different: the frontier AI industry is discovering that models alone do not cross the enterprise gap. That does not make FDEs a failure. It makes them a transitional form. They are what appears before a category has found its true platform layer. SAP does not send SAP employees to every customer This is where the comparison with mature enterprise software becomes useful. SAP does not scale by sending SAP employees into every customer. It has a vast partner ecosystem . Salesforce does not implement every customer itself. It has AppExchange, now evolving into AgentExchange , and a large ecosystem of partners, independent software vendors, and systems integrators. The platform company creates the substrate; the ecosystem industrializes delivery. That distinction matters. When the vendor itself has to supply the scarce human expertise required to make the product work, the category is still immature. When partners, integrators, templates, standards, and repeatable architectures take over, the category begins to scale. This is why the current FDE wave should be read carefully. It is not proof that frontier AI has become a platform. It is proof that it has not yet become one. A true platform reduces the need for bespoke intervention. A preplatform product depends on it. The business model trap There is another problem, and it is more subtle: Once forward deployed engineering becomes a source of revenue, prestige, customer lock-in, and strategic proximity, it becomes harder for the vendor to eliminate it. The very people solving the product’s incompleteness can become part of the business model that depends on that incompleteness. This is classic innovator’s dilemma territory. Clayton Christensen’s argument was that successful companies often struggle not because they fail to see the future, but because their existing business models make the future unattractive or cannibalistic. In this case, the dilemma is simple: If a frontier AI company builds the layer that makes deployments repeatable, modular, and partner-scalable, it may undermine the bespoke, high-touch model that currently brings it close to the largest customers. That is why the real platform may not come from inside the companies training the models. It may come from another layer. The missing layer is not another model The temptation, as always, is to assume that the answer is a better model. A larger model. A more agentic model. A model with longer context, more tools, more memory, more reasoning traces, and more autonomy. But the FDE model suggests something else. If engineers are being sent into customers to map workflows, understand constraints, connect systems, structure context, govern access, and turn AI outputs into operational outcomes, then the missing piece is not simply intelligence. It is architecture. More specifically, it is the layer that turns company reality into something AI systems can operate within: persistent context, process structure, permission models, constraint management, feedback loops, workflow state, business semantics, and outcome tracking. Today, that layer is often reconstructed manually by expert engineers on each deployment. Tomorrow, it will have to become infrastructure. That is the real opportunity. Why this is really BPR with agents This also connects directly to the return of business process reengineering (BPR). In 1990, Michael Hammer’s famous Harvard Business Review article, “ Reengineering work: Don’t automate, obliterate ,” argued that companies should not use technology merely to speed up outdated processes. They should redesign the processes themselves. The idea was right, but in many cases the technology of the time was not yet capable of supporting the ambition. AI changes that, but it also makes the problem more demanding. If companies merely insert AI into existing workflows, they get faster versions of obsolete processes. If vendors merely send engineers to customize each deployment, they get artisanal transformation that does not scale. The real breakthrough comes when the redesign itself becomes systematized: when business processes are not just automated, but represented, governed, adapted, and optimized continuously. That is the point at which enterprise AI stops being a consulting engagement and starts becoming a platform. The FDE is the clue This is why the forward deployed engineer is so interesting. The FDE is not the future of enterprise AI. The FDE is the clue that the future has not fully arrived. The role exists because current systems still require humans to bridge the gap between general AI capability and specific organizational reality. Someone has to translate the company into the machine. Someone has to interpret constraints. Someone has to determine which workflows matter. Someone has to connect data, process, action, and outcome. But history suggests that once a repeatable layer appears, the artisan becomes less central. Web consultants did not disappear after the web matured. But “build me a website” stopped being a mysterious custom engineering problem for most organizations. Enterprise resource planning (ERP) consultants did not disappear after SAP matured. But the ecosystem became standardized enough that the vendor did not need to personally deploy the product everywhere. Cloud architects did not disappear after Amazon Web Services (AWS) became a platform. But infrastructure became programmable, repeatable, and scalable. The same thing will happen here: Forward deployed engineers will not vanish. But if enterprise AI becomes a real platform category, they will become exceptional rather than foundational. The real test of a platform The test is simple: Can the system work without sending the lab? Can it understand the company without a bespoke mapping exercise every time? Can it operate under constraints without manual reconstruction? Can it adapt to workflows without a team of engineers sitting inside the customer? Can partners build on it? Can customers configure it? Can it scale beyond the handful of enterprises that can afford white-glove deployment? Until the answer is yes, we should be honest about what is being sold. It is not AI on tap. It is AI on tap, with plumbers included. And that is fine, for now. Every category has its artisanal phase. The mistake is confusing that phase with the destination. What comes next The next stage of enterprise AI will not be defined by who has the most impressive model or the largest deployment team. It will be defined by who builds the layer that makes those deployment teams less necessary. That layer will not merely answer questions. It will represent the company. It will encode processes, constraints, permissions, memory, and outcomes in ways that AI systems can actually use. It will allow models to operate inside the business rather than hover above it. It will turn bespoke deployment into repeatable architecture. When that happens, the current FDE boom will look obvious in retrospect—not as the final form of enterprise AI, but as the bridge between demos and platforms. And when the real platform layer appears, the industry will change very quickly. Because utilities do not scale by sending engineers to every sink. They scale when the plumbing is already there.
- Etzioni on AI: The Pope can talk, but only we can walk
In a guest op-ed, AI researcher and UW professor emeritus Oren Etzioni responds to Pope Leo XIV's new encyclical on AI, arguing that moral pronouncements — however eloquent — accomplish little unless people change their own behavior. Read More
Score: 42🌐 MovesMay 26, 2026https://www.geekwire.com/2026/etzioni-on-ai-the-pope-can-talk-but-only-we-can-walk/ - In Pictures: 'Phishing as a Service' - Abnormal AI and AUSCERT roundtable
A selection of photos from a recent iTnews roundtable lunch at Uncle Su restaurant at the Star on the Gold Coast.
- Prism Media Launches Prism News, an AI-Native Publishing Platform Operating More Than 200 Niche Publications Nationwide
Prism Media Launches Prism News, an AI-Native Publishing Platform Operating More Than 200 Niche Publications Nationwide USA Today
- RhinOS partners with China’s Keenon on service robot expansion
RhinOS partners with China’s Keenon on service robot expansion 매일경제
- AI is Reshaping the Way Organizations Invest in Their People
Pay must also evolve as AI changes the workplace.
Score: 42🌐 MovesMay 26, 2026https://www.inc.com/maria-colacurcio/ai-is-reshaping-the-way-organizations-invest-in-their-people/91349532 - Corsair's Pro lineup is the company’s answer to the growing demand for AI workstations and servers
The Corsair Pro lineup features various configurations, including Nvidia's Grace Blackwell Ultra (GB300) GPUs.
- EVP Jay Parikh on India and AI’s inflection point
The post EVP Jay Parikh on India and AI’s inflection point appeared first on Source .
Score: 42🌐 MovesMay 26, 2026https://news.microsoft.com/source/asia/2026/05/26/india-and-ais-inflection-point/ - Why most AI agents disappoint in production (and what to fix first)
Why most AI agents disappoint in production (and what to fix first) InfoWorld
Score: 42🌐 MovesMay 26, 2026https://www.infoworld.com/article/4159901/why-most-ai-agents-disappoint-in-production-and-what-to-fix-first.html - What Is a World Model? Inside the AI Idea Behind 2026’s $1 Billion Bet
Everything you need to understand the AI category that everyone is suddenly funding, explained slowly, with sources Hello DataChefs! 👩🍳 In March 2026, a Paris-based AI company you may not have heard of raised $1.03 billion at a $3.5 billion pre-money valuation. It had no product. It had roughly a dozen employees. It had been operating for a few months. The company is called Advanced Machine Intelligence Labs, or AMI Labs (the name is pronounced like the French word for friend). Its co-founder is Yann LeCun , one of the three scientists who shared the 2018 Turing Award . And its entire pitch can be summarized in four words: build a world model. If that phrase felt like it appeared overnight, you are reacting normally. “World model” went from a technical term most data scientists had never heard of to a billion-dollar funding category in roughly six months. I work in physical AI and scientific machine learning, which is a fancy way of saying “AI that has to deal with the actual physical world.” For the last six months, almost every conversation about my research has started with the same question: but what even is a world model? So in this piece, I want to walk you through what a world model actually is, where the term comes from, who is building one in 2026, and what it might mean for your career. No buzzwords. With sources. At a pace you can follow. (PS: this connects to my earlier piece, Model Recovery vs Model Learning . If you have read it, this is the same divide showing up in a much hotter setting.) 🥄 What a world model actually is Let me give you the definition first, then unpack it. A world model is an AI system that learns to predict how the environment around it will change when an agent takes an action. Compare that to what most people already know: A large language model , like ChatGPT or Claude, predicts the next word. A world model predicts the next state of the world. The phrase has a precise academic origin. In a 2018 paper titled simply “World Models,” researchers David Ha and Jürgen Schmidhuber defined a world model as a system that learns a “compressed spatial and temporal representation of the environment.” The agent then uses that internal representation to plan and act. The most famous result in their paper was that they trained an agent to play VizDoom: Take Cover entirely inside its own “hallucinated dream” generated by the world model, and then transferred that learned policy back into the real game, where it scored well above the threshold needed to solve it. That last sentence is the foundation. The agent learned in its imagination. The imagination was accurate enough that the policy worked in reality. The term itself is older. Jürgen Schmidhuber first described this idea in 1990 , in a technical report on using recurrent neural networks to predict future states. The 2018 paper made the idea practical and widely cited. The 2026 funding cycle made it famous. 🍳 The kitchen analogy If that definition still feels abstract, try this. Imagine three different kinds of cooks. Cook number one has memorized a million recipes. You ask her what comes after sautéing the onions, and she tells you, because she has read it written down a thousand times. She is fluent in describing food. But she has never actually cooked. She knows the words, not the warmth of the pan. This cook is a large language model . It is fluent in text about the world without ever having interacted with the world. Cook number two follows a thick rulebook. Page 47 says sugar caramelizes at 320 degrees Fahrenheit. Page 312 says cold butter and warm flour produce flaky pastry. He follows the rules exactly. The food is reliable, but the rulebook had to be written by a human who already understood the kitchen. This cook is a physics simulator , like MuJoCo or NVIDIA Isaac Sim. It models the world correctly because someone hand-coded every law. Cook number three has spent a million hours watching real kitchens. Nobody taught her the rules. She figured them out by observing. Hand her a raw onion and say “chop and sauté.” Before she even lifts the knife, she can mentally play the next thirty seconds: the sizzle, the color change, the smell, the texture of the onion once it softens. She predicts what will happen next, given what she does. This cook is a world model . The rules were learned, not programmed. And the prediction is action-conditioned , which is the technical way of saying her prediction depends on what she chooses to do next. The third cook is what almost every major AI lab is now chasing. 🧂 The kitchen test (how to spot a real world model) The phrase “world model” is being used loosely in marketing copy right now. Here is the simple test I use to tell a real one from a lookalike. A real world model takes in a state (the current situation) and an action (what you decide to do), and gives you back the next state (what happens as a result). The action input is the dividing line. If a system only takes text and produces a video, that is video generation. Beautiful, useful, but not the same thing. If a system takes “this is what the world looks like now, and this is what I will do,” and predicts what the world will look like a moment later, that is a world model. This matters because some impressive video models (like OpenAI’s Sora or Google’s Veo) sometimes get described as world models in casual conversation, but you cannot pause them mid-clip and say “now turn left.” A world model lets you do exactly that. The action is part of the input, not an afterthought. 🍲 The 2026 family tree (one table, everything in it) Here is the table I wish someone had handed me three months ago. I have seen “world model” used to describe at least five different kinds of systems. Only some of them actually qualify. The cleanest takeaway is in the second column. Does the system accept an action as input? If yes, it has the basic ingredients of a world model. If no, it is something else, even if it is impressive on its own terms. 🥘 Who is actually building one in 2026 Here is the short list, as of the date this article was written. AMI Labs . Yann LeCun’s new company in Paris. Building world models using JEPA, an architecture LeCun proposed in 2022 . JEPA stands for Joint Embedding Predictive Architecture, and the short explanation is that it learns abstract representations of how the world changes, rather than trying to predict every pixel. AMI Labs is still in the research phase. NVIDIA Cosmos . A family of “world foundation models” designed for robotics and autonomous vehicles. Companies like Agility Robotics, Figure AI, Waabi, and Uber have adopted Cosmos for training their physical AI systems with synthetic data. Genie 3 . A Google DeepMind system that generates interactive 3D environments in real time at 24 frames per second and 720p resolution. Waymo adopted a specialized version called the Waymo World Model in February 2026 for self-driving simulation. Marble . A commercial product from World Labs, the company co-founded by Stanford computer scientist Fei-Fei Li. Generates 3D scenes you can walk through. Pricing ranges from free to $95 per month. V-JEPA 2 . Meta’s video-based world model, developed by LeCun’s former team at Meta’s FAIR lab. Still being actively published on after his departure. Notice that these systems do not share one architecture. JEPA, diffusion models, transformer-based video models, and 3D scene generators are all sitting under the same label. The category is real. The blueprint is not finished. 🌶️ The honest part Three caveats I think are important, because the marketing tends to skip them. One. Long-horizon coherence is still hard. “Long-horizon” just means the model has to keep things consistent over a long sequence of predictions. If you roll most of today’s world models forward for thirty seconds, objects start drifting, disappearing, or quietly contradicting earlier frames. This is the open research problem, not a finished one. Two. Learning to predict frames that look physical is not the same as understanding physics. A model can produce video that obeys gravity on every test the researchers ran, and still break the moment it sees something outside its training distribution. This is the question I wrote about in Model Recovery vs Model Learning , and it is going to define which world models actually hold up under real-world deployment. Three. Training is genuinely expensive. Frontier-scale world models cost tens of millions of dollars to train and require massive GPU clusters. This is not a category most teams can replicate on their own hardware. Even AMI’s own CEO, Alex LeBrun, predicted that within six months, “every company will call itself a world model to raise funding.” That is your warning label. When that happens, run the kitchen test. 🥄 My personal take (the one no one else is saying out loud) Here is what I actually think will happen. The companies that will own this category in 2027 and 2028 are not going to be the ones with the biggest pure learned models. They are going to be the ones who figure out how to pair learned world models with the physics-aware methods that researchers like me have been quietly building for the last decade. Methods like SINDy and PINNs , Koopman operators, equation discovery, neural ODEs. The reason is simple. Pure learned models are great at imitating physics until they encounter something outside their training distribution, and then they fall apart in ways that are hard to predict. Physics-aware methods are bad at scaling, but they encode constraints that the universe actually obeys. The most useful world models are going to combine both. The companies that figure out that recipe will eat the next decade. If you are a researcher reading this and wondering whether to pivot, that is the lane I would point you to. Not pure end-to-end learned dynamics. Not pure classical physics. The hybrid. 🥢 What this means if you are building an AI career right now If you have spent the last two years learning about LLMs, you have not wasted your time. LLMs are not going anywhere. They are extraordinary at language, code, and reasoning over text, and that capability is being woven into more software, not less. But the next wave of compute, capital, and hiring is moving toward systems that have to reason about physical consequences. That is a slightly different skill set. A few areas that are suddenly worth investing in: Modeling sequences that are not text. Video, sensor data, robot trajectories. Self-supervised learning , especially the JEPA family. The core idea is learning to predict useful representations of future states without needing labels. Reinforcement learning. It quietly stopped being trendy. World models put it back in the center, because the whole purpose of having a world model is to train an agent inside it. Sim-to-real transfer. “Sim-to-real” means taking a policy that was trained in simulation and getting it to work in the real world. This is a hard problem, and it is the bottleneck for most robotics deployments. Physics-aware machine learning. SINDy, PINNs, Koopman operators, and Kolmogorov-Arnold Networks. This is the lane I write in, and the people who have been quietly working in it for a decade are suddenly relevant. If I were starting over today, I would pick one of these and go deep, not wide. The “I touched every framework” résumé is going to age the same way “I tried every JavaScript framework” did in the 2010s. 📖 A small glossary, for later A few terms I used in this piece, defined clearly so you can come back if you forget. Action-conditioned prediction. Predicting the next state given both the current state and the action the agent takes. This is the defining feature of a world model. JEPA (Joint Embedding Predictive Architecture). An architecture proposed by Yann LeCun in 2022 that learns abstract representations of the world by predicting representations of future states, rather than predicting raw pixels. Latent space. A compressed numerical summary of what is in an image, video, or scene. World models usually work in latent space instead of on raw pixels because it is far more efficient. Long-horizon coherence. Whether a model keeps things consistent over many predicted steps. Sim-to-real. The challenge of transferring a model trained in simulation to the real world without it breaking. World foundation model. A large pretrained world model designed to be adapted for many downstream tasks, similar to how GPT-style models are foundation models for language. 🎤 Final Mic Drop If I had to compress this whole shift into one sentence, here it is. 2017’s bet was that AI could see. 2022’s bet was that AI could talk. 2026’s bet is that AI can predict what happens next . That is a harder problem. It is also, honestly, a more useful one. Most of the things we want machines to do, from folding laundry to driving trucks to assisting in surgery, are problems about what happens next, not problems about words. Whether the bet pays off will take years to know. In the meantime, the marketing will get louder, and you will see “world model” used to describe many things that are not quite that. When that happens, run the test. Does it take in a state? Does it take in an action? Does it give back the next state? If yes, that is the real recipe. See you in the next one. 👩🍳 Sources used in this piece Ha, D., & Schmidhuber, J. (2018). World Models. arXiv:1803.10122 . AMI Labs funding coverage. TechCrunch, March 2026 . NVIDIA Cosmos and robotics adoption , NVIDIA Newsroom, 2025. Genie 3 specifications , Google DeepMind. Waymo World Model announcement , Waymo Blog, February 2026. Marble launch and pricing , TechCrunch, November 2025. V-JEPA 2 announcement , Meta AI Blog, June 2025. LeCun, Y. (2022). A Path Towards Autonomous Machine Intelligence . OpenReview. Schmidhuber’s 1990 origin of the term , Wikipedia: World model (artificial intelligence). What Is a World Model? Inside the AI Idea Behind 2026’s $1 Billion Bet was originally published in Towards AI on Medium, where people are continuing the conversation by highlighting and responding to this story.
- The production ceiling: where voice agent stacks start showing their limits
Voice agent stacks' limitations in production
Score: 42🌐 MovesMay 26, 2026https://assemblyai.com/blog/where-voice-agent-stacks-start-showing-their-limits - Tata Elxsi Launches ViTel, a Material Intelligence Solution for Med-Tech Powered by Viridium AI
Tata Elxsi today announced the launch of ViTel™, a Material Intelligence solution for Medical Device manufacturers, co-developed with Viridium AI, a material intelligence platform company. ViTel™ is being formally introduced at DeviceTalks Boston 2026, where Tata Elxsi is presenting its latest advances in AI-powered MedTech engineering and compliance. Medical device manufacturers are facing a new operating reality. Material […] The post Tata Elxsi Launches ViTel, a Material Intelligence Solution for Med-Tech Powered by Viridium AI appeared first on CXOToday.com .
- The agentic divide: Why “good enough” AI isn’t enough to survive the new economy
A widening gap in agent quality is creating a two-tier system where well-resourced firms scale infinitely while small players are trapped by high-friction, "low-trust" tools.
Score: 42🌐 MovesMay 26, 2026https://restofworld.org/2026/ai-agent-inequality/?utm_source=rss&utm_medium=rss&utm_campaign=feeds - At the A.I. Epicenter, Technologists Dismiss Pope Leo’s Warnings About the New Technology
Pope Leo XIV’s spiritual message on artificial intelligence arrived as Silicon Valley’s A.I. enthusiasts pursue their own spirituality through technology.
- What Readers Found When They Asked Their Chatbots About China
Plus, a Hollywood screenwriter says historical disputes sank a U.S.-China film project.
- Freshworks Signals A More Practical Future For AI Service Management
Freshworks’ latest announcements reflect a broader shift in service management AI from feature expansion to operational practicality. The real differentiator is no longer whether a vendor offers generative or agentic capabilities, but whether those capabilities can be adopted quickly, grounded in strong context, and sustained within day-to-day service environments.
Score: 41🌐 MovesMay 26, 2026https://www.forrester.com/blogs/freshworks-signals-a-more-practical-future-for-ai-service-management/ - The ongoing engineering cost of a multi-vendor voice agent stack
Engineering costs of multi-vendor voice agent stacks
- Singapore’s AI paradox: World-class infrastructure, single-digit SME adoption
If you have read anything about AI in Singapore over the past two years, you have probably seen the headlines. World-class AI infrastructure. Top three globally for government AI readiness. Number one in Asia for population-level AI usage. A national AI council chaired by the Prime Minister. Budget 2026 is putting AI at the centre […] The post Singapore’s AI paradox: World-class infrastructure, single-digit SME adoption appeared first on e27 .
Score: 41🌐 MovesMay 26, 2026https://e27.co/singapores-ai-paradox-world-class-infrastructure-single-digit-sme-adoption-20260524/ - The Despair of the Professor in the Age of A.I.
“Was it always the case that half of our students would cheat if it were easy enough?”
Score: 41🌐 MovesMay 26, 2026https://www.newyorker.com/news/fault-lines/the-despair-of-the-professor-in-the-age-of-ai - Ucell and ZTE complete large-scale deployment of AI‑Powered green network solution in Uzbekistan
Network-wide rollout boosts energy efficiency by 10.6%, cutting carbon emissions and operational costs without compromising user experience
- AI's growing footprint raises fire safety challenges for data centers
Over the past decade, the ability of artificial intelligence to speed up complex processes has made it a key tool in engineering research. Many people associate AI with cloud computing, but its growth requires expanding infrastructure, such as data centers.
- AI Use in Marketing Presents Panoply of Legal Peril, Say Advertising Attorneys
"It's just that people are not thinking about the fact that there may be claims embedded in there, in the (AI) content that was created," said attorney Aaron Goodman. "They're just not thinking in the context of having to vet that in the same way."
- Move Over, Tesla. Another Robo-Taxi Stock Is Making Progress.
Move Over, Tesla. Another Robo-Taxi Stock Is Making Progress. Barron's
- Top Australia banker says AI will mean smaller teams, no use ‘pretending otherwise’
Top Australia banker says AI will mean smaller teams, no use ‘pretending otherwise’ The Straits Times
- Observability was built for humans. AI agents need something different
Observability platforms built for humans falter as AI agents demand deeper, continuously accessible telemetry data.
Score: 40🌐 MovesMay 26, 2026https://www.techradar.com/pro/observability-was-built-for-humans-ai-agents-need-something-different - AI Being Used For Therapy And Companionship In Youth And Adults
AI is being used by adults and youth for mental health, companionships, and romantic relationships. We review new data and recommendations from top organizations.