Edition: September 8, 2026

OpenAI claims its latest AI model, more powerful than GPT‑6 Astra, has produced a solution to the Navier‑Stokes equations, one of the seven unsolved Millennium Prize Problems that have eluded mathematicians for 90 years. The breakthrough, announced in a blog post and reported by major outlets, could earn a $1 million prize and marks a historic milestone for AI‑driven scientific discovery.

NYU mathematician Tristan Buckmaster and Anthropic researcher Levent Alpöge claim to have made breakthrough progress on the Navier‑Stokes Millennium Prize problem using OpenAI’s Codex and Claude models, only to discover that OpenAI allegedly raced ahead, producing a full proof with a next‑generation AI after learning of their approach. The dispute raises serious questions about AI‑driven research ethics, credit attribution, and the competitive use of massive compute resources in solving landmark mathematical problems.

OpenAI announced that its AI system has produced a solution to the Navier‑Stokes equation, one of the Clay Millennium problems, claiming the breakthrough required millions of dollars in compute and was formalized in Lean. The claim is contested by mathematicians Tristan Buckmaster and Levent Alpöge, who say OpenAI may have leveraged their unpublished AI‑assisted work and tried to downplay their contributions, sparking a dispute over credit and data use.

DeepMind unveiled AlphaGenome Atlas, a 1‑petabyte resource that predicts the molecular impact of every possible single‑nucleotide variant—about 9 billion changes—in the human genome, and introduces the AVI score to rank variant effects. The freely accessible platform promises to accelerate disease research and biological discovery by providing comprehensive, AI‑driven variant impact data.

French AI lab Mistral AI secured a €3 billion Series D round, valuing the company at over €21 billion, in what it calls the largest equity raise for a European tech firm. Backed by Samsung, a16z, Nvidia and others, the funding will fuel a push for sovereign AI infrastructure, including a 1 GW compute target in Europe and tools that let customers control data locality. The deal underscores growing geopolitical interest in non‑US AI capabilities.

Google DeepMind unveiled AlphaGenome Atlas, an AI-driven predictive map of every possible DNA letter change in the human genome, aiming to accelerate biological research and aid drug discovery. The tool promises to deepen understanding of genetic mechanisms and could speed the development of new disease treatments.

Microsoft's September patch set a new record, fixing about 972 vulnerabilities, including 112 high‑severity flaws, as the industry ramps up defenses against a surge of AI‑driven attacks. Experts warn this spike may become the new normal, underscoring the growing urgency of rapid patching.

Meta launched Muse, a personal AI agent for iOS, Android and web that automates tasks like emailing, booking travel and making purchases, emphasizing built‑in security and privacy through a Secure VM architecture and Stripe’s Link payment protection. The service is free to try but requires a subscription for heavy automation, marking Meta’s push to catch up with competitors while trying to rebuild user trust.

Google Cloud has partnered with Accenture to launch the Accenture Gemini Enterprise Business Group, training up to 1,000 forward‑deployed engineers to help enterprises adopt Google’s AI tools and close the gap with rivals like OpenAI, Microsoft and Amazon. The move underscores the fierce competition among hyperscalers to monetize AI services, as enterprise demand and ROI remain uncertain.

AI researcher Danijar Hafner is launching a stealth startup that builds humanoid robots capable of handling unfamiliar real‑world environments by using model‑based reinforcement learning and world‑model simulations. His Dreamer series of agents have already achieved human‑level performance in video games and now aim to bring that planning‑ahead capability to physical robots for home and other settings.

Geneticist Yuancheng "Ryan" Lu has pioneered a gene‑therapy that reprograms cells to restore damaged optic nerves, reversing vision loss in mice and now entering human trials for glaucoma. The treatment, called ER‑100, uses a trimmed set of rejuvenation genes to avoid cancer risks, marking a potential breakthrough in age‑related eye disease and broader anti‑aging research.

Entrepreneur Laureen Meroueh's startup Hertha Metals has unveiled a novel furnace that turns iron ore into liquid steel in a single step using natural gas instead of coal, cutting emissions by at least 50% and costs by 25%. The pilot plant in Texas can produce a ton of steel per day, and the approach could significantly decarbonize an industry responsible for 7% of global emissions.

Google has launched WeatherNext 3, an AI-powered weather model that delivers hourly, high‑resolution forecasts of wind speed at turbine height, cloud cover and solar irradiance, aimed at grid operators and renewable developers. By offering the data through Google Cloud services, the company enters the lucrative market of AI weather forecasting for energy management, where accurate predictions are critical for balancing renewable generation and rising AI‑driven electricity demand.

Anthropic confirmed that hackers have been stealing Claude tokens by hijacking user session keys and minting unauthorized OAuth tokens, affecting multiple subscribers who saw unexplained usage spikes and charges. The breach, linked to common infostealer malware, prompted Anthropic to invalidate sessions, issue refunds, and warn users, highlighting security gaps in token usage tracking.

Cognition, the AI coding assistant startup behind Devin, secured a $2 billion funding round at a $48 billion valuation, underscoring venture capital confidence that the AI coding market will host multiple dominant players. The company reports rapid revenue growth and aims for $4‑5 billion in annualized revenue by 2026, while grappling with high compute costs and competition from rivals like Cursor, which was recently sold to SpaceX.

OpenAI's latest ChatGPT Images 2.5 update introduces a Sketch tool that lets users draw simple doodles directly in the chat and have the system generate detailed, realistic images from them. By typing @Sketch, users can create a quick sketch, add prompts, and receive a high‑quality AI‑generated picture, expanding how people interact with generative visual AI.

Meta has launched Muse, a personal AI agent that integrates with users' apps and services to handle tasks like emailing, booking travel, and shopping, requiring deep access to personal data and offering free use with paid tiers up to $100 a month. The rollout follows Meta’s recent $18 billion settlement and raises questions about consumer trust and privacy as AI agents move beyond chat to actionable assistance.

Meta unveiled Muse, a personal AI agent that can autonomously handle tasks like shopping, email, and travel planning, aiming to democratize AI and close the gap with rivals such as OpenAI and Google. The move is part of Meta’s multi‑billion‑dollar push to revitalize its AI strategy and bring advanced assistants to a broad user base.

Hannah Earley, CTO of Vaire Computing, is pioneering reversible computing chips that reclaim waste heat as usable energy, a breakthrough that could dramatically boost the efficiency of data centers and personal devices. The startup has demonstrated a resonator that recovers more energy than it consumes, marking a rare practical step for a concept long confined to theory, though commercial viability remains early-stage.

A new paper argues that AI safety guards should target specific harmful subsets of topics rather than blanket‑refusing entire categories, using political persuasion as a case study. The authors introduce a boundary‑aware self‑distillation method to train models that refuse only manipulative content while still answering benign political queries, exposing limits of current self‑generated safety tuning approaches.

A class-action lawsuit has been filed against Anthropic by a group of Claude subscription users who allege the company misrepresented the benefits of its premium Max plan, claiming they received less than promised. The suit, brought by former FTC attorneys, highlights growing consumer scrutiny of AI service pricing and could set precedents for how AI firms market subscription tiers.

Google Chrome has moved to a two‑week release cycle, halving its previous schedule to accelerate security patches and AI‑driven feature updates. The faster cadence aims to shrink the “N‑day” vulnerability gap and keep pace with AI‑generated threats and emerging AI‑enhanced browsers.

Adobe is revamping its Premiere video editor with a new Generative Media tool that lets users generate video clips, sound effects, music and soundscapes directly on the timeline, eliminating the need to switch to external apps. By highlighting empty gaps, editors can summon context‑aware, fully editable AI‑generated assets, streamlining workflow and making the technology more user‑friendly.

Arm has unveiled its Total Design for Physical AI initiative, introducing a new robotics capability framework to standardise terminology and performance tiers across industries such as mining, agriculture and manufacturing. The programme gathers over 80 partners—including AWS, Hugging Face, Siemens and Unitree Robotics—to create common baselines for AI models, hardware, and software, aiming to reduce integration risk and accelerate large‑scale deployment of intelligent robotic systems.

A recent study of 50 vitamin and supplement queries found that YouTube was cited in 53% of Google’s AI‑generated Overviews, making it the most referenced source, ahead of GoodRx and Healthline. The research also revealed that most AI citations point to pages not present in the organic search results, highlighting a divergence between AI summaries and traditional rankings.

Coca‑Cola’s Coke Buddy platform in Malaysia now features an AI‑driven “Perfect Basket” tool that suggests which drinks retailers should order and in what quantities, using data such as past purchases, seasonality and weather. In a recent campaign, over 4,000 outlets tried the feature, with 83% of participants following the recommendations and reporting higher sales growth, though detailed performance metrics were not disclosed.

MIT Technology Review’s latest newsletter spotlights its annual Innovators Under 35 list, naming 35 young scientists and engineers worldwide who are advancing biotechnology, AI, computing, robotics, and climate solutions. The issue also highlights a new study showing spiderwebs can serve as powerful sources of environmental DNA for tracking wildlife.