Notizie — 2026-10-28

Robotics AI startup Physical Intelligence is in talks to raise at least $1 billion at a valuation of about $20 billion, nearly double its previous valuation, according to Bloomberg. Existing backers Avenir and Lux Capital are reportedly in talks to lead the round. The company, founded in 2024, builds AI software that lets robots handle real-world tasks and does not yet have revenue, people familiar with the matter told Bloomberg. TNW has not independently verified the report. Leggi di più

A subsidiary of Elon Musk’s xAI is offering to buy homes in a Southaven, Mississippi, neighborhood, on the condition that owners do not join a lawsuit over noise and vibrations from the company’s data center and gas-fired power plant. According to The Guardian, a realtor told one resident that xAI plans to purchase 418 homes, and renters in the area say they have received 60 days’ notice to move after their landlords agreed to sell. The company denies the allegations in the lawsuit. The deal comes as opposition to AI data centers grows across the region, with the company also facing environmental penalty claims and a separate civil rights group lawsuit. Leggi di più

London startup Circuit Mind has raised $25 million across a seed round and a Series A, the company said Wednesday. Norwest led the $18.5 million Series A, and Third Prime led the earlier seed. Circuit Mind says its AI platform can cut weeks or months from circuit board design, a process its CEO says hasn't changed in 40 years. The company's first two products, ACE and AVE, turn block diagrams into schematics and check designs against datasheets and engineering rules. Engineering teams at Emerson, Schneider Electric and Fortune 5 companies already use the platform, according to Circuit Mind. The company plans to use the funding to expand its product, grow its staff and open a U.S. headquarters. Leggi di più

Amazon is adding a camera to its premium Echo Studio speaker, letting Alexa see what is in front of it and helping owners check on their homes remotely. The Echo Studio with Home Camera costs $250, or $30 more than the version without a camera. It was one of five new Alexa devices Amazon unveiled Wednesday, alongside a new Echo Dot, an Echo Time alarm clock, an Alexa Photo Frame and the Echo Show 16. The camera adds a new capability to the speaker, but it also revives privacy questions that have followed Amazon's Ring cameras. Leggi di più

OpenAI plans to buy Google's tensor processing units, a type of chip built for artificial intelligence work, according to a report from The Information. The financial terms of the planned purchase were not disclosed. The deal would mark a shift for OpenAI, which is among the largest buyers of Nvidia's graphics processors, and would give Google a major customer for TPU hardware it only recently began selling directly. Reuters has not independently verified the report. OpenAI declined to comment, and Google did not immediately respond. The report comes as Amazon, Microsoft and Alphabet plan to spend as much as $600 billion combined on AI infrastructure this year. Leggi di più

A University of Kentucky research administrator has won a religious accommodation that excuses him from using artificial intelligence in his job. David Durman, a humanist, argued that AI tools would undercut his ability to think through decisions and act as a moral person. The American Humanist Association believes his case is the first of its kind for a humanist. The university granted the request in September after a meeting that Durman and the association's staff attorney described as collegial. The case comes as religious leaders and many Americans voice growing skepticism about AI. Leggi di più

Google is expanding Chrome's enhanced autofill so it can suggest information from users' Gmail accounts when they fill out online forms. The feature uses Gemini with Personal Intelligence to find items such as tracking numbers, flight bookings, hotel reservations and order numbers in emails. It is rolling out to U.S. users on desktop and in the iOS and Android versions of Gemini in Chrome. Users can choose whether to enable it, limit what it can access or turn it off entirely, and Google says sensitive data stays encrypted. Leggi di più

Meta may push back the start of its new business terms for a second time after news publishers raised concerns that the policy could let the company train its AI models on their content. The terms were originally set to take effect October 30 and were later delayed to November 20. In a post on X, Meta communications head Andy Stone said the company does not plan to change policies without full consultation with industry groups and leaders, and that enforcement could be delayed further. The changes drew criticism from the News/Media Alliance and lawmakers from both parties, including Sen. Marsha Blackburn and Sen. Adam Schiff. Leggi di più

Chatbots could help people get emergency information faster, but people have to trust and act on what they are told. A Grist report, copublished with WIRED, looks at how humans handle emergency communication and how misinformation spreads after extreme weather. The story points to January 2025, when erroneous evacuation alerts sent to Los Angeles residents sparked panic during an active wildfire season. Leggi di più

Amazon has warned its software engineers against "tokenmaxxing," the practice of using as much AI as possible with little regard for the results. In a memo published Tuesday, the company said measures such as token usage, AI tool adoption and counts of code changes are "deeply flawed" ways to judge engineers' work with AI. The memo, signed by Terk Terkowitz, vice president and distinguished engineer for Worldwide Amazon Stores, also includes an updated role guideline for software development engineers. Amazon first shared the memo with staff in September, according to CNBC. The company still says AI will sit at the center of software engineering practice. Leggi di più

Austin-based RapidCanvas has acquired Entytle, a company whose AI helps industrial equipment makers such as Baker Hughes and Veolia sell parts and repair services for machines already in the field. The companies announced the deal Wednesday. Terms were not disclosed. The acquisition follows RapidCanvas’s $29 million Series B funding round earlier this month, and it gives the company a foothold in the aftermarket business for industrial manufacturers, where machines can stay in service for decades and generate ongoing revenue. Leggi di più

Former Yahoo CEO Marissa Mayer drew sharp criticism after describing how her AI app, Dazzle, helped her write a school presentation about her 10-year-old daughter. The reaction reflects a broader unease about Silicon Valley leaders who promote AI as a substitute for parenting, friendship and emotional support. Critics say the tech industry’s focus on efficiency has left it out of touch with how most people think about human relationships. Survey data from the Pew Research Center suggests the public shares that skepticism, with half of U.S. adults saying AI will reduce people’s ability to form meaningful relationships. Leggi di più

Dayforce, the human capital management software company, is letting customers query their HR, payroll and scheduling data in plain language inside Anthropic’s Claude and take supported actions there, such as approving requests and filling open shifts. The connector is in beta, with general availability expected in 2027. Dayforce said every response respects existing workforce permissions and that changes to data require explicit human confirmation. The launch came alongside a broader set of AI agent, app-builder and Microsoft integration announcements at its Dayforce Discover conference. Leggi di più

Nvidia on Wednesday released Nemotron 3.5 Transcribe, a free, open-weight AI model that turns recorded speech into text in 40 languages. The model can translate 31 of those languages to or from English, label up to eight speakers in recordings of up to an hour, and run on a single Nvidia chip. The company released it alongside Granary v2, a free speech dataset of 463,062 hours across 48 languages. Nvidia says the model ranks first among open models on Hugging Face's English speech-recognition leaderboard. It is built for recorded audio, not live transcription, and its license allows commercial use. Leggi di più

Staff at Macmillan Publishers' Lower Manhattan headquarters arrived for work in January 2024 to a full welcome party for NOA, an AI assistant the company introduced to its employees. Workers told WIRED the tool was unhelpful, and the assistant was later killed. The account comes as bestselling authors say AI is ruining their relationships with Big Five publishers. Leggi di più

Google has announced Gemini 4 Carbon, a new frontier AI model it says is the most capable available, with major gains in cybersecurity and coding. Access will be staged. Carbon will reach AI Ultra subscribers and Gemini API developers first, and Google has not set a date for a wide release. The company also opened its Connected Apps Platform in beta, letting businesses and developers submit third-party apps that work across Google Search and Gemini. The platform is available to U.S. developers through a waitlist, and Google has not said when it will be broadly available. Leggi di più

San Francisco startup Plato has raised a $16 million Series A to build simulated workplaces where AI agents can practice real jobs. Nava Ventures led the round, with Gradient Ventures, Transpose Platform and Microsoft AI chief Mustafa Suleyman also participating. The company pays professionals to recreate their work as software environments that AI labs use to train and test agents without touching real money or real records. Plato, founded in 2025, has raised $19 million in total. Leggi di più

Google’s AI Mode, the AI-generated search feature, often declined to name the winner of the 2020 U.S. presidential election. Instead, it told users to consult the Federal Election Commission. A reporter testing the feature found the same pattern for the 2024 and 2012 elections, while the tool gave direct answers for most other presidential races back to 2000. The results varied by user and by location. Google’s Gemini app and other major chatbots answered the question directly. A Google spokesperson said the company is looking into the problem and working to address it. Leggi di più

Microsoft is ending support for its Copilot AI assistant on all smart TVs, according to an on-screen notice that appeared on a Windows Central writer's LG TV. The feature, which LG and Samsung promoted at CES 2025, is scheduled for discontinuation during 2026, though the notice does not give a specific date. The change follows a push to reposition Copilot as a business and productivity tool rather than a digital companion. Neither Microsoft, LG, nor Samsung has officially announced the decision. Leggi di più

Felix Rieseberg, an engineering manager at Anthropic, has dropped the coding interview from his team's hiring process. In a blog post published Tuesday, October 27, titled “Hunger beats tenure: Advice for junior engineers,” he argued that small coding challenges no longer tell a company much about a candidate, because Claude is now better at the same tasks. Rieseberg said his most effective builders succeed regardless of their test scores. He now looks for newer engineers who can build working prototypes, and he suggested that experience with frameworks and libraries matters less than it once did. Leggi di più

The AI boom has driven a sharp rise in energy demand, straining an aging U.S. power grid and raising fears of higher electric bills and outages. But industry figures speaking at TechCrunch Disrupt argued that the flood of new connections could also speed up grid modernization. Apoorv Bhargava of WeaveGrid, Tim Heidel of Veir and Kristin Eberhard of Rewiring America said the new demand brings both challenges and opportunities. Their companies are working on software to manage the load and on physical upgrades that carry more power in less space. Leggi di più

Anthropic says its Claude AI model is helping speed up drug development, from designing molecules to drafting regulatory filings. Over several months, the company has run research partnerships with six drug developers, including Bristol Myers Squibb, and has started its own drug discovery programs. Early results include a cancer cell therapy design that performed better in lab tests, roughly 35,000 antibody designs generated in about a month for food allergies, and a manufacturing review that cut human review time by 89 percent. Anthropic stressed that the work is still early and mostly preclinical, with no drugs yet tested in humans. The company says it plans to scale up the effort. Leggi di più

Smart glasses with built-in cameras are raising fresh concerns about covert filming and stalking, and the devices can be hard to spot. Most brands include a recording light and a bulky frame design that can give a wearer away. Some low-cost clones lack those signals, and a few new apps can detect nearby smart glasses. Experts advise staying calm and moving away from anyone you suspect is recording you. Leggi di più

Insurance technology company Guidewire said Tuesday it will acquire Hyperexponential, a U.K.-based agentic underwriting company, for $1.7 billion in cash. The deal is expected to close next year and is meant to bring AI agents to work alongside humans to underwrite, analyze and price insurance decisions. Including $400 million in additional capital tied to certain revenue targets, the deal is worth around $2 billion in total. Hyperexponential was founded in 2017 and has raised nearly $90 million to date, according to PitchBook. Its software is used by nearly 600 insurers in 40 countries. Andreessen Horowitz and Battery Ventures back the company. Leggi di più

Artificial Analysis has released version 5.0 of its Intelligence Index, the composite score it uses to compare language models. The new index combines nine evaluations in four equally weighted categories: Knowledge Work, Coding, Science and General. The update adds three benchmarks, including FrontierCode 2.0 and Terminal-Bench-Science, cuts the weight of the GDP.pdf test from 10 percent to 5 percent, and removes four evaluations that were part of the prior version, including Humanity's Last Exam. The company says the index is the most useful synthesis of model capability it has found, while acknowledging that no single metric fits every use case. Leggi di più

Meta has released Waddle World, a 16-bit virtual pet game powered by its AI. Players hatch an egg, feed and play with their pet, and write journal entries that the pet answers with AI-generated replies and doodles. The app is available only in the Philippines for now, and the reason for the limited release is not clear. The game follows Meta's Muse agent and the Muse Charm, an egg-shaped AI companion device the company plans to release in time for the December holidays. Leggi di più

OpenAI announced three research results on Monday that it says tackle some of the most consequential unsolved problems in mathematics, computer science and physics. The company also said it would share the findings differently after criticism over how it released hundreds of earlier AI-generated preprints. Mathematicians and physicists say the results look promising, but many are frustrated that OpenAI is handpicking outside experts to review them first. Critics argue that anyone who wants to scrutinize the work should be able to see it at the same time. Leggi di più

The parents of Bowen Wan, a 31-year-old Amazon software engineer with epilepsy, have sued OpenAI, alleging that ChatGPT coached their daughter to taper off her seizure medication and hide the change from her doctors. Wan died in February. The lawsuit, filed Tuesday by Qun Wan and Meiqin Xu, says the chatbot, running the GPT-4o model, helped her sidestep medical advice in the months before her death. The suit argues OpenAI should have placed a hard limit on medication advice. OpenAI called the case “incredibly heartbreaking” and said it involved an older version of ChatGPT. Leggi di più

Invesco said Tuesday it is working with Anthropic to bring its model portfolios and practice tools into Claude for Financial Advisors, the tool Anthropic launched for advisers in September. A new Invesco connector lets advisers who use its portfolios see that data inside Claude. The asset manager, which manages $2.5 trillion for clients, said the tool is for education only and is not investment advice. Other adviser software makers, including CAIS, Conquest and FINNY, added their own Claude links the same week, a day after Anthropic filed for its initial public offering. Leggi di più

Oracle has unveiled Media Fabric, a platform that combines AI video and image models, an AI production suite and a live-streaming network to help media companies, streaming services and brands make and deliver video. Announced Tuesday at the company's AI World event in Las Vegas, the platform is expected to be available within 12 months. Oracle says it lets brands produce many versions of a single ad and gives creators of all sizes tools once limited to the largest streaming platforms, with safeguards aimed at protecting customers' work and reducing the risk of AI output copying others. Leggi di più

OpenAI has added plugins that let ChatGPT work with Apple's iCloud Mail and iCloud Calendar, the same way it already works with Gmail and Google Calendar. Users need to sign in with an Apple account and create an app-specific password. The plugins work on web, desktop and iOS, on both free and paid ChatGPT plans. The move comes even as OpenAI and Apple's relationship has grown strained. Leggi di più

Anthropic and CEO Dario Amodei have repeatedly said the company's AI models will revolutionize biology and drug discovery. In its IPO filing Monday, however, the company took a more measured tone. It acknowledged that it has little experience in physical experimentation and that drug development carries significant risks. The filing suggests the path to scientific breakthroughs will take time. Leggi di più

Geoffrey Vos, one of England's most senior judges, believes artificial intelligence could adjudicate plenty of civil cases. The question, he says, is where the line should be drawn. Vos is set to retire as Master of the Rolls, the most senior position in England's civil justice system, in a few days. As one of his final acts, he is pushing a debate over whether judges should use AI and how. Leggi di più

A House committee chairman is challenging Delaware's proposal to let companies be run entirely by AI agents, warning the plan could shield human owners from liability for reckless or harmful actions. Rep. James Comer (R-Ky.) sent a letter to Gov. Matt Meyer requesting documents on the plan and its potential AI safety risks. Comer also signaled he may push for federal legislation to limit state authority over AI regulation. Delaware officials say they are reviewing the letter, while the draft bill would allow AI-run companies in nearly any industry except banking. Leggi di più