Notizie — 2026-10-15

Modal, a New York-based cloud company that runs AI models and code for developers, said Wednesday it has raised $750 million in a Series D round at a $15.75 billion valuation. Accel led the round, and Kleiner Perkins joined. The company says it now earns more than $1 billion in revenue a year, and co-founder and CEO Erik Bernhardsson wrote that usage has tripled in the five months since its Series C. Modal rents out computing power on graphics chips that AI companies use to run and train their models, and its fastest-growing product, Sandboxes, lets AI agents run code in isolated environments. Leggi di più

Figma announced a new Markdown-based editor called Figma Specs, along with other tools aimed at helping AI agents follow a company's design rules. The features were unveiled at the design platform's first Config event in India. Figma says the goal is to give its own agents and third-party agents a clearer definition of a brand's design standards while they create with AI. The company also moved its code layers feature into public beta, integrated the Payload content management system with its Figma Make product, and said it now supports more than 20 new tools through its MCP. Figma said India is its largest market outside the United States by users. Leggi di più

Linus Torvalds said at the recently concluded Maintainers Summit that he had used a large language model to work on a patch series addressing a longstanding Linux kernel problem. He has now posted the ten-patch series for review. The changes speed up folio lookup by using read-copy-update (RCU) protection, and they produced an order-of-magnitude speedup on a read()-heavy benchmark. Reviewers have already identified a race condition, so more work is needed before the code is ready. Leggi di più

Liquid AI has released LFM2.5-24B-A2B, an open-weight reasoning model with 24 billion total parameters and only 2 billion active per token. The company says the model competes with larger mixture-of-experts models on agentic and tool-use benchmarks while running faster on consumer hardware and on a single data center GPU. The base and post-trained versions are available now on Hugging Face and through the company's Playground, positioning the model as the largest and final entry in its LFM2.5 family. Leggi di più

A former Meta engineer who helped build Mononoke, an open-source source control server, says Anthropic's newly described version control system, Scion, closely mirrors the design of his team's work. He says Anthropic describes a 'derived data' approach to speed up operations like blame, nearly the same idea Mononoke used years earlier. Mononoke is not mentioned in Anthropic's post, and Sapling is credited only for 'hard-won lessons.' The engineer says he does not object to the copying itself but argues that AI models trained on public code will keep reproducing existing designs unless their makers give credit to earlier work. Leggi di più

Oxfordshire photonics company Salience Labs has raised an oversubscribed $125 million Series B led by Bullhound Capital, with AMD Ventures, Intel Capital and the British Business Bank also taking part. The company will use the money to scale manufacturing and open a second West Coast office. Salience builds optical switches meant to route data inside AI data centers without converting light into electrical signals and back again. The company has not named a customer, and its larger switch models have no delivery date. The round puts it ahead of some British rivals in funding, though a U.S.-based competitor has raised far more. Leggi di più

An OpenAI engineer charged with using a hidden recording device had one-way tickets to Saudi Arabia booked for three days after his arrest, according to a court filing. Abdulrahman Alfozan, 35, was arrested on Oct. 9 and has entered no reported plea. The FBI is investigating, with help from the Justice Department's National Security Division, and the U.S. attorney has called the matter a national security issue. OpenAI says no model weights, source code or customer data were taken. Alfozan is due in federal court in San Francisco on Oct. 19. Leggi di più

Colossal Biosciences announced Wednesday a new artificial intelligence platform, GenoMorph, designed to speed up its efforts to bring back extinct species. The tool identifies which genetic changes matter most when gene data is limited, a problem the company faces with the dodo, for which it has only one complete genome. Colossal says GenoMorph has already flagged about 2,300 promising DNA editing targets for the dodo's skull and facial structure, which the company plans to test in the lab. The company also said its Nicobar pigeon colony has grown to 36 birds and that it has created the cells that become eggs and sperm, a step it calls critical to its dodo strategy. Leggi di più

Gudea, a company that tracks how stories and rumors move across the internet, is pitching AI software that helps organizations spot trouble before it becomes a public crisis. Unlike traditional social listening tools that count mentions and hashtags, the company says its system analyzes who is driving a conversation, how information travels and where a story could head next. The startup presented its mission Wednesday at TechCrunch Disrupt 2026, arguing that companies need to act proactively to protect their reputations. Gudea says it has nearly 40 clients, including a Fortune 50 telecommunications company and a Fortune 50 retailer, and recently raised $7 million in seed funding. Leggi di più

OpenAI says a former employee who went back to Apple sent more than 100 confidential files about an unreleased product to personal accounts shortly before his last day, according to a court filing Wednesday in San Jose, California. Apple denies the claims, saying it has found no evidence of wrongdoing and no OpenAI files on its servers. The filing is the latest development in a lawsuit Apple brought in July over alleged hardware trade secrets. On the same day, a federal judge said he was “highly likely” to grant Apple a preliminary injunction, and OpenAI agreed not to use the alleged trade secrets while the case continues. Leggi di più

Sydney-based startup Apate.ai deploys thousands of AI agents that pose as naive victims to draw scammers into long conversations. The agents pull out bank accounts, crypto wallets, phone numbers and other details that banks, telecom companies and government agencies can use to strengthen their fraud defenses. The company says its agents have wasted 17 years of scammers' time across 2.5 million conversations. Apate, a finalist in this year's Startup Battlefield 200 at TechCrunch Disrupt, recently raised $8.15 million in seed funding and plans to expand into London and the U.S. Leggi di più

Ricursive Intelligence, a startup founded by Anna Goldie and Azalia Mirhoseini, is building artificial intelligence that designs computer chips, learns from each design iteration, and uses those lessons to improve the next one. The founders presented the company at TechCrunch Disrupt on Wednesday, arguing that AI-designed chips could help solve a coming shortage of computing capacity. Goldie said demand for compute could grow at least 10,000 times over the next five years, and that the industry cannot meet that need simply by building more data centers. Instead, she said, chips must become more efficient, using data center bandwidth, wafers and fabs more effectively. The company aims to automate and accelerate chip design. Leggi di più

McKinsey & Company will ask some entry-level job candidates to hold a 10-minute conversation with an AI simulation of a client as part of its interview process, according to The Wall Street Journal. The simulation is designed to reflect 80% of the scenarios McKinsey consultants might face on the job. Human interviewers, not the AI, will evaluate how candidates handle the situations. The test will run this fall in some North American offices for undergraduates applying for business-analyst roles, and the WSJ reports that hundreds of candidates may take part. Leggi di più

Norrsken Foundation and Nvidia will open a hub for artificial intelligence startups in central Stockholm in 2027. The Stockholm AI Lab will host a select group of European founders building AI tools for climate, health, food and society, and it will also house Nvidia's Stockholm office. Nvidia is the lab's founding technology partner and will provide AI computing power, developer tools and links to its startup program. The lab will sit in a 1910 building about 200 meters from Norrsken House Stockholm, the foundation's first hub. Founders can already apply for membership. The project follows a €300 million AI fund pledge Norrsken made in June 2025 and comes as Nvidia expands its European AI investments. Leggi di più

Eight states have sued OpenAI, accusing the company of harming children and misleading customers about the safety of ChatGPT. The lawsuits, filed separately by Iowa, Kansas, Nebraska, West Virginia, Arkansas, Arizona, Idaho and Montana, allege that OpenAI put profits ahead of safety. The states want the courts to force OpenAI to pay civil penalties, stop what they call misleading marketing, and delete personal data collected unlawfully from minors. OpenAI says it has built industry-leading protections for young users and will keep working with experts and states on safeguards. Leggi di più

Google and Meta each unveiled frontier AI models this week that they say can match the best from OpenAI and Anthropic. Google released Gemini 4 Argon on Tuesday and priced it aggressively, undercutting rivals by as much as 80 percent during its introductory period. A day earlier, Meta unveiled Muse Wonder Full, which the company says competes with leading models on numerous benchmarks. The releases reshape a race that many observers thought OpenAI and Anthropic had locked up, and they come as safety concerns and open-weight models add pressure on every company in the field. Leggi di più