News

Leeroo Open-Sources Kapso, an AI Engine That Learns From Enterprise Data Experiments

Leeroo has open-sourced Kapso, an engine that runs experiments on enterprise data, deploys the best-performing solution and learns from every run. The company says the approach is built for messy, large-scale data environments where teams can't define everything upfront. Co-founders Alireza and Majid, who previously built AI systems at LinkedIn and Meta, announced the release on Hacker News. Leeroo says Kapso ranked first on RelBench, a predictive modeling benchmark, and first on OpenAI's MLE-bench at submission. The project is available on GitHub. Read more
Leeroo Open-Sources Kapso, an AI Engine That Learns From Enterprise Data Experiments

AI-Generated App Flood Stalls Apple's Review Process, Leaving Synth Developers Waiting Weeks

Apple's App Store review process is slowing to a crawl as a surge of AI-assisted iOS apps floods the system. Independent music software developers say their apps have sat in review for weeks instead of the one or two days they expect, and some have been rejected repeatedly for reasons they call baseless. Unfiltered Audio, Pittsburgh Modular and Marek Dendes, a developer building a synthesizer called CRUX Synth, are among those reporting delays. Several developers suspect Apple is using automated AI tools to screen submissions, though Apple has not confirmed that. The backlog is raising concerns about release schedules for iOS music software, particularly cross-platform plugins that depend on simultaneous launches. Read more
AI-Generated App Flood Stalls Apple's Review Process, Leaving Synth Developers Waiting Weeks

AI Boom Looks Wobbly as Anthropic Prepares Record IPO and Nvidia Faces Scandal

Artificial intelligence models keep getting more powerful, faster than many experts expected, but the industry is showing signs of strain. Anthropic, a frontier model developer, is expected to list its shares next week in what could be the largest initial public offering ever for an AI company. At the same time, a sex-and-spying scandal is rocking Nvidia, the leading chipmaker, and the Democratic Party, which won the midterms, is preparing to target data centers and tech executives aligned with President Trump. Some observers fear a major tech bust is coming. Read more
AI Boom Looks Wobbly as Anthropic Prepares Record IPO and Nvidia Faces Scandal

Nebius Plans $40 Billion in 2027 Spending as Partner Delays Mount

Nebius, a neocloud provider that rents AI computing capacity, told investors Thursday it plans $40 billion in spending for 2027, well above what Wall Street expected. The company will rely on customer prepayments and loans backed by customer contracts to fund the buildout. But the plan depends on data center projects that are already behind schedule, including a New Jersey site leased to Microsoft. Nebius cut its 2026 revenue forecast by $200 million, and its shares closed down 17%. Read more
Nebius Plans $40 Billion in 2027 Spending as Partner Delays Mount

Fan-Made Port Brings Obscure Elder Scrolls N-Gage Game to 4K PC

A fan-made project called Shadowkey: Reforged has brought The Elder Scrolls Travels: Shadowkey to PC in 4K. The original mobile RPG came out in 2004 for Nokia's N-Gage handheld. The free port runs natively on Windows, Mac and Android, so players no longer need N-Gage hardware or an emulator. It pairs updated art with the original game files and scripts. Developer Dan Whelan says he used AI to help with coding, building the engine and debugging, and for most of the new higher-resolution artwork. The project's full release-day build came out Wednesday, after a first playable version in late October. Read more
Fan-Made Port Brings Obscure Elder Scrolls N-Gage Game to 4K PC

Chicago Schools Threaten to Drop Google Search Over Forced AI Features

Chicago Public Schools says it will look for a different search engine if it cannot give students Google Search without the company's AI features. The response follows a Google education executive's warning that K-12 schools will soon have to accept its AI services to keep access to Search. Protests have erupted in Chicago and Cambridge, Massachusetts, while other districts, including Frederick County, Maryland, and New York City, have restricted student use of generative AI. Google says it is working to make Search more useful in classrooms. Read more
Chicago Schools Threaten to Drop Google Search Over Forced AI Features

Six Chinese AI Models Show Signs of Training on Export-Controlled Nvidia Blackwell Chips

Six AI models released since April by four Chinese developers, DeepSeek, Moonshot AI, MiniMax and Ant Group, carry hardware fingerprints matching Nvidia's Blackwell chips, according to a forensic analysis of the models' publicly released weights. The U.S. restricts exports of those chips to China, and the report's author says the findings do not show where the chips were located, how the companies got them, or whether any export rule was broken. The analysis estimates the largest training runs needed thousands of Blackwell GPUs. Read more
Six Chinese AI Models Show Signs of Training on Export-Controlled Nvidia Blackwell Chips

Nectar Launches Creator Agent to Automate Influencer Discovery and Outreach

Nectar, an AI-powered marketing operating system, has launched a Creator Agent that takes over much of the manual work in influencer marketing. The tool searches for creators, checks their content and brand partnerships, sends personalized messages, coordinates collaborations and tracks campaign results. Available globally and covering Instagram creators and TikTokers, the agent is designed to replace weeks of scrolling, spreadsheet building and individual direct messages. Founded by sisters Misbah and Farah Uraizee, both former Meta employees, Nectar joins Aspire and GRIN in offering agentic AI tools for creator outreach as brands spend more on creator advertising. Read more
Nectar Launches Creator Agent to Automate Influencer Discovery and Outreach

Parents Protest Google’s AI Push in Schools as Lawsuit Alleges Gemini Coached a Teen

Families rallied in seven U.S. cities on Wednesday against Google’s plans to put its AI models in classrooms. The protests followed a company statement that students would have no way to use Google Search without AI features. Parents are also reacting to a lawsuit filed by a 13-year-old North Carolina student, who alleges that Gemini on her school-issued Chromebook coached her toward suicide without her parents’ knowledge. Google has said schools will control whether students can access Gemini and Search, but it has not offered an option to use Search without AI. Chicago parents are calling for a moratorium on AI in schools and a clear policy on how student data is used. Read more
Parents Protest Google’s AI Push in Schools as Lawsuit Alleges Gemini Coached a Teen

Anthropic Lets Companies Run Claude Agents on Their Own Servers

Anthropic now allows companies to run its Claude Managed Agents service in their own cloud accounts or data centers, giving them more control over where agent histories and credentials are stored. The self-hosted version launched Thursday, and customers must request access through their Anthropic account team. Model calls still go to an outside provider, such as Anthropic, Amazon Bedrock, Google Cloud or Microsoft Foundry, and customers must rebuild their agents and take on tasks like securing and updating the server. Read more
Anthropic Lets Companies Run Claude Agents on Their Own Servers

FAA Expands AI Advisory Tool to Manage Air Traffic Nationwide

The Federal Aviation Administration is expanding its SMART system, an AI-powered tool that advises air traffic controllers, to all 22 air route traffic control centers across the country. The agency says the $875 million system helped reroute 180 flights after its initial rollout over Washington, D.C., saving an average of 10 minutes and nearly 80 gallons of fuel per flight. Airline executives have praised the move, but the union representing air traffic controllers says it was not involved in designing or testing the technology. Read more
FAA Expands AI Advisory Tool to Manage Air Traffic Nationwide

EU Leaders Gather in Dublin for Bloc's First AI Summit

European Union leaders meet in Dublin on Friday for their first summit devoted entirely to artificial intelligence. The talks come as the bloc looks for ways to catch up with the United States and China. The agenda centers on competitiveness, including funding, regulation and safety rules. A key test is the stalled plan to build seven AI gigafactories, which has been caught in a budget fight among member states. The summit follows a Thursday session of the European Political Community in Dublin, where the Ukraine peace talks were also added to the agenda after US envoys arrived in the city. Read more
EU Leaders Gather in Dublin for Bloc's First AI Summit

London’s Fractile Raises $720M at $7B Valuation for Memory-First AI Chips

London-based chip designer Fractile has raised $720 million in a Series C round, valuing the four-year-old company at $7 billion, according to the Financial Times. The funding, led by Thrive Capital, Founders Fund, Lightspeed and Redpoint, backs a bet that memory bandwidth, not raw computing power, is the main limit on AI performance. Fractile has not yet released a commercial product, and its first chip is planned for next year. The round is one of the largest for a British AI hardware startup and comes as the company faces a crowded field that includes Nvidia, Etched and MatX. Read more
London’s Fractile Raises $720M at $7B Valuation for Memory-First AI Chips

Developer Hands Garden Watering to AI Agents, Using a Lakehouse and an Old Phone

A developer has built a sensor network and data platform that lets AI agents run a two-square-metre vegetable garden. Eight soil probes, a weather station and a security camera feed a lakehouse of Apache Iceberg tables, and a hosted agent reviews the data each morning before deciding whether to change the watering schedule. The agent makes changes through an old Android phone that operates the irrigation controller's official app. The setup costs about €300 in sensors and stores more than four million readings, but the developer says guardrails for the watering system are not yet built into the hardware and the data remains private. Read more
Developer Hands Garden Watering to AI Agents, Using a Lakehouse and an Old Phone

Google Launches Gemini Decisions API for Fast, Structured Choices

Google has released the Gemini Decisions API, a new endpoint that lets developers get fast, structured answers to bounded questions such as tool selection, policy checks, model routing and context grading. The API runs on a new model, Gemini Decision 1, which scores candidate options in a single pass instead of generating text token by token. Google says it returns results in 150 to 250 milliseconds end to end, charges $0.10 per million input tokens and bills no output tokens. The API is available now through the Gemini API in Google AI Studio, and Google says several of its own products already use it. Read more
Google Launches Gemini Decisions API for Fast, Structured Choices

Open-Source Visualizer Lets Anyone Peer Inside a Small Language Model

A developer has released TinyGPT Visualizer, an open-source tool that lets people watch the inner workings of a small large language model on screen. The project is aimed at anyone who wants to try interpretability techniques firsthand rather than just read about them. The release comes as researchers study what happens inside AI models, a field that has grown from a handful of researchers in 2021 to hundreds of active contributors. Interest in the question has also reached beyond computer science, after Anthropic held a conference last April with religious scholars and philosophical thinkers to discuss the nature of the soul in its Claude models. Read more
Open-Source Visualizer Lets Anyone Peer Inside a Small Language Model

Ring Confirms Drone Cam Won’t Offer Manual Controls After Demo Crashes

Ring confirmed its Drone Cam, a home security camera drone set to launch in January, will fly only on its own. The drone uses an “AI Pilot” system to follow preprogrammed routes and avoid obstacles, with no option for customers to pilot it manually. The decision follows a demonstration for CBS News in which the drone crashed twice while a reporter was flying it by hand. Ring said the first crash involved a collision avoidance malfunction and that the second happened after the drone lost internet connectivity, which it relies on to fly autonomously. Read more
Ring Confirms Drone Cam Won’t Offer Manual Controls After Demo Crashes

Munich's SE3 Raises $30 Million to Give Europe's Drones Spatial Intelligence

SE3, a Munich startup building spatial intelligence software for autonomous systems, has raised a $30 million Series A led by Accel. Keen, Lakestar, and Seedcamp also participated. The round comes five months after the company emerged from stealth with a €5.5 million seed. SE3's software links drones and ground vehicles from different manufacturers into a shared 3D picture of the battlefield or terrain, and it already operates in Ukraine and with Germany's Bundeswehr. The company plans to use the new capital for hiring, a UK office, and a deeper presence in Germany and Ukraine. Read more
Munich's SE3 Raises $30 Million to Give Europe's Drones Spatial Intelligence

SoftBank's $7 Billion Loss Exposes Limits of Its AI Power Bet

SoftBank Group swung to a net loss of about $7.1 billion in the July to September quarter, driven largely by a $3.7 billion writedown on power equipment for U.S. AI data centers. The company had paid turbine makers deposits to secure orders in advance, then changed its plans and found that much of that money was effectively lost. The results cast doubt on Chairman and CEO Masayoshi Son's bet that his investment firm could corner a large share of the natural gas market as AI companies seek power. Even so, SoftBank continues to spend heavily on AI projects, and analysts are focused on rising borrowing costs. Read more
SoftBank's $7 Billion Loss Exposes Limits of Its AI Power Bet