1. DeepMind Open-Sources WeatherNext Cyclone Models, Claims More Than a Day of Additional Forecasting Lead Time Google DeepMind has opened its WeatherNext cyclone-forecasting models to researchers and weather organizations, releasing code and model weights alongside a Nature paper.

2. SoftBank Donates $50 Million to Trump Presidential Library, Announces Federal-Land Data Center Project Two Months Later SoftBank said it donated $50 million in January to the Donald J. Trump Presidential Library Foundation.

3. ChatGPT to Remove Text-Only Chat Limits for Free and Go Users Next Week OpenAI says ChatGPT users on its free and Go tiers will be able to hold unlimited text chats starting next week.

In Brief

• Google Maps Adds Food Ordering, Hotel Search, and Ticket Links Google is rolling out new Ask Maps capabilities in the U.S. for ordering food through supported platforms, comparing hotel availability, and finding event tickets. Personal Intelligence, disabled by default, can use Gmail and Google Calendar data to personalize answers in markets where Ask Maps is available.

• Cloudflare Open-Sources Its AI App-Building Workspace Cloudflare released Cloudflare OS, an internal platform that lets employees create applications from natural-language descriptions, as open-source software. Its security model runs individual app instances in separate V8-based isolates and manages access permissions centrally.

• Naïve Raises $28.5 Million for Agent-Operated Business Infrastructure Naïve raised a $28.5 million Series A led by Nexus Venture Partners after signing up more than 30,000 developers for its business-provisioning API. The platform lets AI agents arrange services such as company incorporation, email, payments, cloud infrastructure, and storage, while retaining human checks for identity verification and sensitive actions.

• Mirendil Signs Google Cloud Compute Deal Worth More Than $100 Million AI research startup Mirendil signed a multiyear Google Cloud agreement giving it access to TPUs, Nvidia GPUs, and managed training clusters. CEO Behnam Neyshabur said the deal is worth upward of $100 million and will support research into systems designed to improve their own knowledge and performance iteratively.

• Omilia Raises $67 Million to Expand Its Customer-Service Automation Athens-based Omilia raised a $67 million Series B led by Expedition Growth Capital to expand its voice and customer-support platform. The company, which reports $60 million in annual recurring revenue, plans to open another U.S. office, strengthen sales operations, and grow from roughly 500 to 600 employees this year.

• Suno Plans Watermarks and Tighter Download Rules for AI Music AI music generator Suno said it will introduce watermarking and fingerprinting technology intended to identify its generated tracks and combat fraud and misuse. The company also plans to revise its download policy but has not disclosed the policy’s final details.

• OpenAI and the APA Partner on Youth Mental-Health Guidance OpenAI and the American Psychological Association, a professional organization for psychologists, announced work on evidence-based guidance, resources, and safeguards concerning young people’s use of AI.

• Former Spotify Engineers Bring Intent Prediction to Online Stores Malachyte raised $10 million in seed funding to expand an e-commerce personalization system built by three former Spotify employees. Its software updates product recommendations from shoppers’ real-time behavior and has been generally available to Shopify merchants since June 2026.

• Florida County Imposes One-Year Data-Center Moratorium Commissioners in Hernando County, Florida, unanimously approved a yearlong pause on data-center construction amid organized local opposition. Residents cited concerns including groundwater, chemicals, and the broader effects of AI infrastructure development.

• Genome Models Produce Functional Bacteria-Infecting Viruses Stanford researchers used large genome models to generate genomes for viruses that infect bacteria, producing variants closely related to an existing virus but with distinct features. The researchers said the results warrant preparation for the possibility that future systems could design viruses targeting vertebrates.

• Benchmark Finds Persistent-Memory LLMs Invent User Attributes MirageBench evaluated 12 models and found that each inferred unsupported user attributes in 35% to 49% of assessed claims. Model self-assessments were unreliable for comparing systems, although external judging showed stronger agreement with blind human annotation.

Read the full edition →