AI News Digest

1. [10/10] Worried Anthropic researchers warn that AI ‘could kill all humans’

Source: The Verge | Published: 2026-09-09T05:56:28-04:00 (4h ago)

Google Cloud's partnership with Accenture to deploy forward-deployed engineers signals a major push to accelerate enterprise AI adoption and integration.

A senior Anthropic safety researcher has said there is more than a 10 percent chance artificial intelligence "could kill all humans" by the end of the decade, just hours after a colleague resigned over fears the AI lab and its rivals are carelessly racing to build "superhuman systems" they cannot control. In a post on X announcing his departure, Jacob Coxon, a researcher who has trained AI systems at Anthropic, said he had quit the company over its lax approach to safety. Coxon, who previously trained systems for OpenAI, accused the two AI companies of "racing straight to self-improving superintelligence and gambling with our lives," even … Read the full story at The Verge.

https://www.theverge.com/ai-artificial-intelligence/991927/anthropic-ai-kill-all-humans

2. [10/10] Google Cloud races to catch up in the AI deployment wars with Accenture deal

Source: TechCrunch | Published: Tue, 08 Sep 2026 16:20:31 +0000 (22h ago)

Google Cloud's partnership with Accenture to deploy forward-deployed engineers signals a major push to accelerate enterprise AI adoption and integration.

Google Cloud expands its enterprise AI push with Accenture, betting on forward-deployed engineers to drive adoption and overcome deployment bottlenecks.

https://techcrunch.com/2026/09/08/google-cloud-races-to-catch-up-in-the-ai-deployment-wars-with-accenture-deal/

3. [10/10] Coca-Cola uses AI to improve retailer ordering in Malaysia

Source: AI News | Published: Tue, 08 Sep 2026 10:00:00 +0000 (1d ago)

Coca-Cola's implementation of AI to optimize retailer ordering in Malaysia provides a concrete example of AI-driven business process automation and supply chain efficiency.

Coca-Cola is using AI to recommend which products Malaysian retailers should order and in what quantities through its Coke Buddy platform. The Perfect Basket feature uses Coca-Cola’s Central Recommendation Engine to analyse previous orders, ordering frequency, seasonality, weather, and purchasing patterns among similar businesses. Coca-Cola said Coke Buddy currently supports about 39,000 retail outlets across Malaysia. The company describes Coke Buddy as a self-ordering platform that allows retailers to buy products through its app, website, or WhatsApp, with personalised order suggestions and order tracking also available. Perfect Basket builds on those existing ordering functions by recommending both products and quantities before a retailer completes an order. Retailers can review the recommendations and retain control over what they purchase. How Perfect Basket guides retailer orders Coke Buddy already uses previous purchase history to suggest products a retailer is likely to order again. Perfect Basket adds other signals, including seasonality, weather, ordering frequency, and purchasing trends among comparable businesses. Perfect Basket recommends products and quantities before retailers submit their orders through Coke Buddy. Fulfilment is handled separately by Coca-Cola Refreshments Malaysia or its suppliers under existing sales and distribution arrangements. Coca-Cola said its sales teams remain involved with retailers alongside the digital ordering system, while retailers retain control over the final purchasing decision. Coca-Cola recently disclosed usage figures for Perfect Basket following its Perfect Basket, Perfect Ride campaign, which ran from January to April 2026. The campaign encouraged retailers to use the recommendation feature when placing orders and received more than 4,500 entries from over 4,000 retailers in Malaysia. During the campaign, 83% of participating outlets adopted Perfect Basket recommendations, according to Coca-Cola. The figure applies only to retailers taking part in the campaign, not the full network of about 39,000 outlets supported by Coke Buddy. Coca-Cola also said participating outlets that followed the recommendations recorded higher sales revenue growth than comparable retail outlets. The company did not disclose the size of the difference or provide detailed performance data showing how individual recommendations affected sales or inventory levels. The available Malaysian campaign data does not provide figures for forecast accuracy, stock availability, inventory levels, or logistics costs. Suggested orders extend beyond Malaysia Coca-Cola has deployed similar suggested-order capabilities elsewhere in its bottling network. In its first-quarter 2024 results, the company said it and its bottling partners had connected nearly eight million customers to B2B platforms globally, while AI-enabled suggested-order capabilities had reached more than three million outlets in Latin America. Coca-Cola has said these systems combine customer data, external information, and AI to generate predictive order recommendations. Then-chief executive James Quincey said in 2024 that digital ordering also allows retailers to adjust deliveries without waiting for a salesperson to visit. Coca-Cola has also linked suggested orders to changes in its sales process. Quincey said AI-generated orders allow pre-sales staff to spend less time taking routine orders and more time on account development, while retailers continue to make the final purchasing decision. Coca-Cola has reported results from earlier pilots using similar recommendation systems. In its second-quarter 2024 earnings call, the company said retailers receiving AI-generated product recommendations based on previous orders and market data were more than 30% more likely to purchase the recommended SKUs in initial pilots. These results did not relate specifically to Perfect Basket in Malaysia. In a separate demand-prediction project, Coca-Cola combined historical sales data with weather and geolocation information to generate replenishment recommendations. CIO Neeraj Tolmare told Fortune in 2025 that a three-country pilot recorded sales 7% to 8% higher than outlets that were not using the AI algorithm. Perfect Basket remains available after the campaign. Coca-Cola said it plans to continue developing Coke Buddy and the recommendation feature using retailer feedback and data, while retailers will continue to have access to the company’s sales representatives alongside the digital ordering system. (Photo by Mahbod Akhzami) See also: MG Ship adds AI route optimisation as logistics returns accelerate Want to learn more about AI and big data from industry leaders? Check out AI & Big Data Expo taking place in Amsterdam, California, and London. The comprehensive event is part of TechEx and is co-located with other leading technology events including the Cyber Security & Cloud Expo. Click here for more information. AI News is powered by TechForge Media. Explore other upcoming enterprise technology events and webinars here. The post Coca-Cola uses AI to improve retailer ordering in Malaysia appeared first on AI News.

https://www.artificialintelligence-news.com/news/coca-cola-ai-retailer-ordering-malaysia/

4. [9/10] Drama swirls around OpenAI’s legendary mathematical milestone

Source: The Verge | Published: 2026-09-08T16:53:52-04:00 (17h ago)

Mistral's €3 billion funding round solidifies its position as a major player in the AI ecosystem, influencing future model access and sovereign AI strategies for global enterprises.

OpenAI says it found a solution to a major math problem that has remained unsolved for around 90 years, as reported earlier by The New York Times and Wired. In a blog post on Tuesday, OpenAI announced that it discovered a solution to the Navier-Stokes problem - which relates to the flow of liquid and gas - using an internal AI model more powerful than the newly released GPT-6 Astra alongside 10,000 concurrent agents. The Navier-Stokes problem is one of seven Millennium Prize Problems, each of which comes with a $1 million reward for solving. OpenAI says it started training the internal AI model on August 28th, which has "exhibited unpreced … Read the full story at The Verge.

https://www.theverge.com/ai-artificial-intelligence/991710/openai-navier-stokes-solution

5. [9/10] Meta bets on AI agent Muse to catch up in AI race

Source: The Verge | Published: 2026-09-08T15:00:00-04:00 (19h ago)

This discussion on AI safety and refusal mechanisms provides crucial insights for enterprises designing robust AI governance frameworks and advanced prompt engineering strategies.

Meta is making another push to bring artificial intelligence to the masses with Muse, a personal assistant it says can put AI in the hands of virtually anyone. The product is the latest step in a multi-billion-dollar strategy overhaul designed to revitalize the company's ailing position in the AI race and help it catch up to rivals like OpenAI, Anthropic, and Google. Muse is a "personal AI agent" designed to help out with everyday tasks and projects, like online shopping, sending emails, and planning a trip. Once given a goal, Meta says Muse can work on its own, opening a browser, filling out forms, and even negotiating on users' behalf. F … Read the full story at The Verge.

https://www.theverge.com/ai-artificial-intelligence/991216/meta-bets-on-ai-agent-muse-to-catch-up-in-ai-race

6. [9/10] Safety for Whom? Refusing the Right Subset of a Topic, Not the Whole Topic

Source: Hugging Face | Published: Tue, 08 Sep 2026 14:23:07 GMT (23h ago)

This discussion on AI safety and refusal mechanisms provides crucial insights for enterprises designing robust AI governance frameworks and advanced prompt engineering strategies.

https://huggingface.co/blog/MultiverseComputingCAI/safety-for-whom

7. [9/10] Mistral raises €3B as sovereign AI becomes big business

Source: TechCrunch | Published: Tue, 08 Sep 2026 14:17:48 +0000 (1d ago)

Mistral's €3 billion funding round solidifies its position as a major player in the AI ecosystem, influencing future model access and sovereign AI strategies for global enterprises.

The French AI lab has raised €3 billion at a €21 billion valuation in a Series D round led by Samsung, Scaleup Europe, and PSG Equity.

https://techcrunch.com/2026/09/08/mistral-raises-e3b-as-sovereign-ai-becomes-big-business/

8. [9/10] AI weather forecasting enters the energy market as Google targets grid operators with WeatherNext 3

Source: AI News | Published: Tue, 08 Sep 2026 09:00:00 +0000 (1d ago)

Google's WeatherNext 3 model extends AI weather forecasting to the energy market, offering specialized AI-powered insights for grid operators.

Google’s newest AI weather forecasting model predicts wind speed at 100 metres above the ground, roughly the height of a modern wind turbine. It also forecasts cloud cover and how much sunlight reaches the surface, and it updates every hour. Energy traders, grid operators and wind and solar developers already pay other companies for that data. The introduction of WeatherNext 3 now puts Google in their market. Google DeepMind and Google Research released the model on September 3. It produces a global forecast every hour at up to five-kilometre resolution for surface variables such as temperature and moisture. The previous version, WeatherNext 2, worked on a 25-kilometre grid and refreshed every six hours. Google says the new energy variables are meant to help grid operators and developers predict how much power their wind and solar assets will generate, then match that against demand. The consumer side of the launch has had most of the attention. WeatherNext 3 now powers weather results in Google Search, the Gemini app, Google Maps and the Google Maps Platform Weather API. Behind it sits an enterprise layer that matters more commercially. The same forecast data can be queried in BigQuery and Earth Engine or downloaded in bulk from Google Cloud Storage, with no model setup required by the customer. Why the energy sector is buying AI weather forecasting Grid operators are running a system that has become harder to predict at both ends. On the generation side, renewables now account for most new capacity. S&P Global Market Intelligence’s US Grid Outlook 2026 projects solar and energy storage as the primary sources of new capacity this year, at 51.2GW and 25.7GW respectively out of more than 90GW of planned additions.  Solar and wind generate according to the weather rather than demand, so each gigawatt added makes a short-term forecast more accurate. On the consumption side, the new load is coming from AI. S&P Global identifies the spread of data centres across North America as a primary driver of the recent surge in electricity demand, forcing utilities to revise their load forecasts upward. Deloitte’s 2026 Power and Utilities Industry Outlook projects peak demand growing by roughly 26% by 2035, with data centre demand alone potentially reaching 176GW, five times its 2024 level. The cost of getting a forecast wrong is straightforward. If an operator underestimates how much wind power will arrive, it has to buy replacement electricity at short notice, usually from gas plants kept on expensive standby. If it overestimates, wind and solar farms end up being paid to switch off because the grid cannot absorb what they are producing. Both outcomes are expensive, and both are forecasting failures. The market Google is entering Selling weather forecasts to the energy sector is an established business. Vaisala, Solcast, DNV’s WindGEMINI and IBM’s HyperWatch all compete in it. So does Jua, a Swiss firm that claims its EPT-2 model beats Microsoft Aurora and DeepMind’s earlier GraphCast on accuracy while updating 24 times a day, against what it describes as a typical four updates a day among competitors. Google’s advantage is reach. The same forecast appears as a table in BigQuery, a layer in Earth Engine, an API in Google Maps Platform and the default answer in Google Search. No specialist vendor has that spread, and the hourly refresh closes the update-frequency gap those vendors have used to differentiate themselves. The incumbents have one technical argument left. Jua’s published position is that physics-based models such as ECMWF’s HRES still outperform purely data-driven AI models during record-breaking extreme weather, because physics models encode rules about how energy and mass move through the atmosphere, while AI models learn patterns from past data.  Jua sells a physics-constrained product, so the claim serves its own interests. It also describes the conditions grid operators worry about most, when a storm falls outside anything the model has seen in training. What is new, and what is being oversold WeatherNext 3 system architecture showing satellite mosaic and analysis inputs producing gridded forecasts, station data and cyclone tracks. Photo from Google’s blog The architectural claim behind WeatherNext 3 is that it learns from real observations instead of from simulations. Most AI weather models, WeatherNext 2 included, are trained on output from numerical weather prediction models, which are supercomputer-driven physics simulations that carry a six-hour data lag. That lag can introduce bias in fast-changing variables such as rain and surface temperature. WeatherNext 3 ingests live geostationary satellite imagery and trains directly on readings from individual weather stations. The shift is real, though narrower than much of the coverage has suggested. Google’s own system diagram shows the model taking in one-hour satellite mosaics alongside traditional historical analysis. DeepMind senior research scientist Ilan Price told Bloomberg the gain comes from not waiting for the next analysis and using the most recent information available instead.  Reporting puts the remaining data lag at three to four hours, down from about seven. Dependence on numerical weather prediction has been reduced, not removed. The accuracy figures need similar care. Google reports improvements of up to 60% against NASA’s IMERG satellite product, 30% against MRMS radar and 10% against rain gauge readings at early lead times, measured using a standard scoring method for probability forecasts. Those are three separate baselines, and the percentages do not add together. The widely repeated claim of 50% better precipitation forecasting applies specifically to forecasts a day or more ahead. Every figure carries an “up to” qualifier, which makes each one a best case rather than a typical result. Google published no independent third-party validation alongside the launch. It points instead to live evaluations by Brightband, whose leaderboard it cites in claiming WeatherNext 3 is the most accurate global weather model to date. A utility considering a switch away from a paid specialist will care more about performance in its own service territory, on its own assets, than about a global leaderboard position. Google’s own stake in the problem Google is selling forecasting tools into a grid problem its own industry helped create. The data centre build-out driving the load growth utilities are struggling to serve is led by the hyperscalers, Google among them, and Google has signed multi-gigawatt renewable procurement agreements to supply its own facilities. Accurate prediction of wind and solar output is directly useful to a company matching large volumes of clean energy against a load that is both growing and variable. That is commercial logic, and it goes some way to explaining why the energy variables shipped in this release. Google has not published pricing for enterprise access to WeatherNext 3, or said whether the BigQuery and Earth Engine data carries standard Cloud query charges or a separate licence. Utilities weighing a move away from a paid specialist will want that figure before they weigh any accuracy claim. 2025, more than 65,000 employees in its Corporate and Investment Bank were actively using the platform, while more than 90% of its engineers were using AI coding assistants. The bank also said AI-based transaction screening allowed it to review more than twice the previous transaction volume while reducing manual operator checks by half. Bank of America is using a generative AI-enabled system called EricaAssist with more than 18,000 customer service employees. The tool summarises why a customer is calling, retrieves relevant information, and recommends possible next steps while keeping the employee responsible for the interaction. Bank of America said in July 2026 that EricaAssist can deliver contextual guidance in under three seconds and has reduced average call times by nearly one minute. The bank plans to extend the system to additional servicing scenarios and business lines later in 2026. (Photo by Google) See also: MIT AI forecasts extreme weather without historical data Want to learn more about AI and big data from industry leaders? Check out AI & Big Data Expo taking place in Amsterdam, California, and London. The comprehensive event is part of TechEx and is co-located with other leading technology events, click here for more information. AI News is powered by TechForge Media. Explore other upcoming enterprise technology events and webinars here. The post AI weather forecasting enters the energy market as Google targets grid operators with WeatherNext 3 appeared first on AI News.

https://www.artificialintelligence-news.com/news/ai-weather-forecasting-google-weathernext-3-energy/

9. [6/10] Sequoia doubles down on Cymphony as AI agents create new enterprise security risks

Source: TechCrunch | Published: Wed, 09 Sep 2026 13:00:00 +0000 (1h ago)

This profile details an entrepreneur's work on AI agents capable of advanced planning, a key development in future agentic workflows for enterprise settings.

Cymphony was valued at more than $100 million in a $25 million Series A co-led by Sequoia and SMBC Fin Atlas Beyond Fund.

https://techcrunch.com/2026/09/09/sequoia-doubles-down-on-cymphony-as-ai-agents-create-new-enterprise-security-risks/

10. [6/10] Students who use AI generally score worse at school

Source: The Verge | Published: 2026-09-09T08:05:01-04:00 (2h ago)

Adobe is enhancing Premiere Pro's AI generators to be more intuitive, simplifying creative workflows for video editors.

Students who use AI to help them study tend to perform worse at school than those who don't, according to data from a global OECD educational report. The situation is more complex than it sounds though, with certain types of AI use giving learners a slight boost, especially among students taught to critically assess how well the AI tools perform. The OECD's Programme for International Student Assessment (PISA) takes data from countries around the world every few years. This year's study, based on data collected in 2025, is the first to be carried out since AI use went truly mainstream. It tests 15-year-old students in science, math, and rea … Read the full story at The Verge.

https://www.theverge.com/ai-artificial-intelligence/991956/student-ai-use-scores-oecd-pisa

11. [6/10] AI power users claim Anthropic duped them with subscriptions, and they’re taking it to court

Source: The Verge | Published: 2026-09-08T13:27:31-04:00 (20h ago)

Arm's new Total Design for Physical AI and robotics framework establishes hardware and software standards for embedded AI applications.

Anthropic says power users are key to its business - it's prioritized them even when it means cutting off other popular applications, like OpenClaw. But some of these same customers say Anthropic misled them into believing they'd get more out of a top-tier pricing subscription than they did. In an expanded class action lawsuit filed today, a group of Claude subscribers say the company deceptively advertised the limits of its Max subscription tier. The lawsuit was brought by attorneys Monica Vaca and Kati Daffan, who both formerly worked at the Federal Trade Commission under Lina Khan. It's a rare attempt to legally penalize AI companies fo … Read the full story at The Verge.

https://www.theverge.com/ai-artificial-intelligence/990313/anthropic-class-action-lawsuit-pricing-subscription-plans

12. [6/10] Adobe is trying to make its AI generators idiot-proof in Premiere

Source: The Verge | Published: 2026-09-08T09:00:00-04:00 (1d ago)

Adobe is enhancing Premiere Pro's AI generators to be more intuitive, simplifying creative workflows for video editors.

Suspenseful clock ticking… as you wait for AI to take your job. | Image: Adobe Adobe is overhauling how editors interact with AI in its Premiere professional video editing software. Its new Generative Media tool makes it easier to generate video, sound effects, music, and soundscapes without ever leaving the project timeline. The generators themselves aren't entirely new; the big change is how easily you can access them without breaking your workflow. Instead of jumping between browsers or external apps to find missing B-roll or audio elements, editors can now highlight an empty gap on the video or audio track and generate context-aware, fully editable clips without leaving the project. Editors can choose from multip … Read the full story at The Verge.

https://www.theverge.com/tech/991133/adobe-is-trying-to-make-its-ai-generators-idiot-proof-in-premiere

13. [6/10] Arm launches Total Design for Physical AI and robotics framework

Source: AI News | Published: Tue, 08 Sep 2026 11:44:10 +0000 (1d ago)

Arm's new Total Design for Physical AI and robotics framework establishes hardware and software standards for embedded AI applications.

Arm has launched Arm Total Design for Physical AI alongside a new robotics framework to establish common standards across automated systems. Physical industries – spanning mining, agriculture, manufacturing, and global transport – account for trillions of dollars in economic activity and an estimated $200 billion annual compute opportunity by the 2030s. To address engineering fragmentation across these sectors, Arm is convening more than 80 partner organisations spanning software, hardware, and AI. Initial ecosystem participants include AWS, ECARX, Hugging Face, Liquid AI, NXP, PlusAI, PSYONIC, QNX, Qwen, Siemens, and Unitree Robotics. The initiative targets physical systems that combine AI models, runtime software, compute silicon, sensors, and actuators to sense, reason, and act in operational environments. Hardware manufacturers and software developers require standardised baselines to reduce integration risk, optimise compute workloads, and move from proof-of-concept testing to deployment at scale. Arm standardises capability tiers for robotics systems Robotics currently lacks a common method to describe, compare, and communicate system capabilities, according to an architectural manifesto (PDF) published by Arm chief architect Richard Grisenthwaite. This fragmentation makes robotic systems harder to design, integrate, and scale across industrial deployments. In response, Arm has introduced the Robotics Capability Framework as a collaborative starting point for a shared technical vocabulary, patterned after the SAE Levels used for driving automation. Arm’s new framework categorises robotic systems across progressing tiers of operational sophistication, mapping machines from reactive setups to context-aware, cognitive, and self-improving systems. Each capability tier links real-world use cases to machine behaviours, outputs, and hardware constraints. These criteria establish parameters for system latency, compute placement, memory allocation, power constraints, determinism, and safety standards. Arm developed the initial baseline using feedback from across the robotics sector. Participating organisations contributing to the framework include Anaxi Labs, ANYbotics, FMC³ Robotics, Fourier, GALBOT, Gravis Robotics, Lenovo, McKinsey, and Robotec.ai. Virtual platforms accelerate pre-silicon automotive physical AI development Arm Total Design for Physical AI extends a collaborative development structure previously used for cloud AI infrastructure. The programme brings together AI models, virtual platforms, digital twins, sensors, compute silicon, and software stacks to enable earlier development and testing cycles. Autonomous transport and robotics face common technical requirements across sensory perception, AI processing, real-time control, safety, and power-efficient compute. Arm demonstrated this collaborative methodology in the automotive sector alongside AWS, Google, HERE, RemotiveLabs, and Siemens. The participating automotive companies developed an integrated digital cockpit reference solution. This environment enabled software engineering teams to develop, test, and validate complex automotive code on the Arm Zena CSS platform prior to physical silicon availability. Arm is now soliciting technical contributions from the wider engineering community to expand the Robotics Capability Framework as physical AI implementations progress. Learn more about physical AI during the Physical AI Expo held in Amsterdam, London, and North America. See also: NVIDIA Jetson Orin Nano 2 brings physical AI to drones and robots Want to learn more about AI and big data from industry leaders? Check out AI & Big Data Expo taking place in Amsterdam, California, and London. The comprehensive event is part of TechEx and is co-located with other leading technology events including the Cyber Security & Cloud Expo. Click here for more information. AI News is powered by TechForge Media. Explore other upcoming enterprise technology events and webinars here. The post Arm launches Total Design for Physical AI and robotics framework appeared first on AI News.

https://www.artificialintelligence-news.com/news/arm-total-design-for-physical-ai-and-robotics-framework/

14. [6/10] This AI entrepreneur is developing agents that can plan ahead for the unexpected

Source: MIT Tech Review | Published: Tue, 08 Sep 2026 10:34:00 +0000 (1d ago)

This profile details an entrepreneur's work on AI agents capable of advanced planning, a key development in future agentic workflows for enterprise settings.

Danijar Hafner’s office in San Francisco’s SoMa district sits mostly empty. His brand-new startup is still in stealth mode and doesn’t even have its name on the door. On the day I visit, there’s only one other person there, and little in the way of furniture. But what it lacks in decor, it makes up for in robots. Humanoids of various shapes and sizes hang like marionettes from racks that run down the center of the wide-open space. While Hafner, 31, won’t say too much about his new venture just yet, he describes it as a continuation of his longtime work to enable AI to navigate environments it has not encountered in training. The humanoids, which he imports from China, are the next evolution of this work—and its physical embodiment. Their ability to react in previously untested scenarios will be key to getting robots into human spaces. Because if you want to send a robot into a person’s home, for example, it needs to be able to handle a floor plan and furniture it’s never seen before.  To achieve this, Hafner relies on something called model-based reinforcement learning. He develops world models—AI models designed to emulate physical reality—and trains agents within them. The agent essentially treats the model as a real-world simulation and learns how to act there. It then uses those experiences to make predictions (to dream or imagine, Hafner might say) about future outcomes. That allows agents—or the robots they’re embedded in—to navigate unfamiliar situations IRL. “I get to interact with a lot of really smart people in research at Google, and he easily sits in the top half of 1%.” Timothy Lillicrap, Google DeepMind Unlike other efforts, Hafner’s technique enables agents and the robots they control to execute massively complicated tasks without the real-world trial-and-­error training that’s traditionally been used in robotics.  Hafner grew up in a rural town in northeastern Germany, where his parents were both classical musicians. He learned programming from a neighbor, and in high school he began taking online courses about AI, which quickly developed into a passion. “I was always fascinated with how thinking works,” he says. AI offered him a way to emulate it on a computer. In 2015, as a second-year under­graduate studying engineering at Hasso Plattner Institute in Potsdam, he won a role as a student researcher at Google Brain. From there, he went on to a dozen internships and other positions at the company, including stints with Google Brain and Google DeepMind (the two have since merged under DeepMind) in the UK, Canada, and the US. He worked with industry legends including Geoffrey Hinton, who is often referred to as one of the godfathers of AI, and Ashish Vaswani, coauthor of the groundbreaking research paper “Attention Is All You Need,” which described the transformer technology used by today’s large language models. One of Hafner’s former managers and coauthors at Google, Timothy Lillicrap, describes him as a standout among standouts. “I get to interact with a lot of really smart people in research at Google, and he easily sits in the top half of 1%,” Lillicrap says. “In many cases he would build, single-­handedly, things it would take entire teams of engineers to build.” Over the years, Hafner has honed and proved his approach by pitting agents trained within his world models against popular video games. His first breakthrough was PlaNet, a model that allowed agents to execute actions by planning ahead. His Dreamer 2 was the first agent to hit human-level performance playing Atari 2600 games using a world model. Dreamer 3 was the first one to solve the Minecraft Diamond challenge—successfully mining in-game gems on its own. And Dreamer 4 went a step beyond that by learning to mine diamonds from an offline data set of recorded game-play videos, without ever interacting with the game directly.  More recently, he’s begun to migrate his agents out of the virtual world and into physical reality. His DayDreamer project used the Dreamer algorithm to let robots operate themselves in novel environments and react to new experiences (such as being pushed over) without any specific training.  Today, Hafner is working on his new startup, which he left Google DeepMind to form in the fall of 2025. Though he’s coy about his next steps, it’s clear he’s dreaming big: “I was interested in solving a problem,” he hints, “that would change the world.” 

https://www.technologyreview.com/2026/09/08/1142088/danijar-hafner-developing-plan-ahead-agents/