Actualités — 2026-11-11

Applied mathematicians Steven Strogatz and Alex Townsend argue in their new book, Big Math: The Hidden Codes That Run Our World, that computers have moved beyond serving as tools for mathematicians. In an interview with Ars Technica, Strogatz said AI systems now produce results that people cannot explain, a shift he describes as moving from servant to scout to oracle. The change raises questions about certainty, understanding and the sense of discovery that drew many people to the field. Strogatz said he sees both loss and promise, and that AI may eventually become a useful teacher. He also said some mathematicians are grieving. Lire la suite

Microsoft has released Automatic Super Resolution (Auto SR), an upscaling technology that competes with Nvidia's DLSS, for the Xbox Ally X, Xbox Ally X20, and select Windows Copilot Plus PCs. The full release works in handheld mode and on battery power, not just when docked, and supports games running at 1080p. Microsoft is also adding an Auto SR Performance mode that can reach up to 120 frames per second and new compatibility with Auto HDR. The performance mode and Auto HDR support will arrive through Windows updates. Lire la suite

Apple is bringing its Siri AI camera mode to iPads as part of iPadOS 27.3, and users willing to run beta software can try it now. The feature builds on Visual Intelligence, letting users ask Siri questions about whatever the camera sees. iPads get an extra option that iPhones lack: Apple Pencil support, which lets users circle an object in a photo to learn more about it. The public release is expected in December, but the first beta arrived in early November. Users who want it sooner can install the beta through Settings, though Apple's guidance is to back up the device first because beta software is more prone to bugs and app problems. Lire la suite

Seven months after Anthropic launched Project Glasswing, an independent tracker counts 442 vulnerabilities (CVEs) attributable to the company's AI models. That is about 0.5% of all CVEs published this year, and only two have been exploited in the wild. Anthropic's public ledger shows nearly 38,000 findings but just 217 fixes, with duplicates and withdrawn reports piling up. A batch of reports sent to the curl project has so far produced six confirmed low-severity bugs out of 34 published reports. The analysis questions whether the project's claims match the evidence. Lire la suite

Meta is pushing a software update to its AI glasses after some devices became “too warm” when moisture got inside them. The company calls the problem rare and says it affects only a small number of glasses. Owners of Meta-branded frames are being urged to update to version 130, and anyone affected can get a free replacement by contacting Meta Customer Service. Meta says Oakley and Ray-Ban models are not affected. The company has not said how many pairs are involved. Lire la suite

Claude now lets teams brainstorm on shared whiteboards and build spreadsheets with the AI, keeping work inside the chatbot instead of in separate programs. The company is launching Claude Whiteboards and Claude Sheets in beta and making its Claude Dashboards product generally available as Claude Data. The three tools are available on all paid plans, and teams can share their work and edit it together. The update is aimed at cutting down on copying output between apps, since Claude previously lacked live access to changes made in those other tools. Admins on Team plans get all three turned on by default, while Enterprise customers have to switch them on, with Claude Data set to turn on by default on November 19, 2026. Lire la suite

Product design workflows were built for an era when writing code was the slowest step, according to designer Vitalii Khomenko, principal designer at Orizon Design and founder of Fluente LLC. Now that AI-assisted engineering has sped up development, he argues, design has become the constraint. His proposed fix is to keep Figma for exploration and deliver final work as code, so engineers can build from components directly instead of re-creating screens from design files. He also says designers will need deeper technical skills while keeping their brand, motion and marketing range, and that teams should bring design judgment in earlier. Lire la suite

Apple appears to be developing a new Mac app called Harbor that would turn a Mac into a private AI inference server, according to leaker Aaron Perris, who found evidence of the software. The app would let users run open-weight AI models on Apple silicon without sending data to the cloud. Harbor can pool the memory of multiple Macs connected over Thunderbolt 5, and it is currently in private preview under Apple's Claris brand. Perris's findings suggest a public launch could come in the weeks and months ahead. Lire la suite

Sahir Raoof, a former SignalFire investor and healthcare startup executive, has raised $27.5 million for his first solo venture capital fund. His firm, Proximal Ventures, will back pre-seed and seed-stage startups that use AI to transform the tools, infrastructure and services drug developers rely on. Raoof argues that pharmaceutical companies have long outsourced much of their work to service providers, and that AI could let startups deliver those services faster and cheaper. The raise comes as first-time fund managers face one of the toughest fundraising environments in years. Proximal has already backed seven startups, and its backers include SignalFire, Kleiner Perkins and Bertelsmann Healthcare Investments. Lire la suite

Audiences are relying on generative AI systems for information at a growing rate, according to a new Reuters Institute survey of people in nine countries, including the United States. Thirty-eight percent said they used AI to get information in the last week, up from 24 percent in 2025. Yet public comfort with AI-made journalism remains low. Only 12 percent said they were very or somewhat comfortable with news made entirely by AI. The findings, detailed by researcher Felix Simon, suggest AI is becoming a major gateway to information even as people stay skeptical about its role in newsrooms. Lire la suite

Google is extending its Gemini-powered Rambler dictation tool to the Pixel 10 series, giving owners of the older flagship phones a feature previously tied to the Pixel 11. Rambler turns spoken rambling into clean, grammatically correct text, removing filler words and smoothing out self-corrections. The Pixel 10 rollout also brings Gemini Intelligence features such as personalized suggestions and multi-step automation for tasks like ordering groceries or booking a taxi. Google's November Pixel Drop adds further AI tools across several device families, and its Pixel Tag tracker is now available to buy. Lire la suite

A once-fringe idea that much of the internet is run by bots and AI-generated content is gaining traction as automated traffic grows. A 2026 report from cybersecurity group Thales found bots accounted for 53 percent of all web traffic, up from 51 percent the year before. The theory traces back to internet forums in the late 2010s and a 2021 post on Agora Road. Its believers say the web feels empty of people, and researchers point to disinformation, election risks and a growing backlash against AI as reasons for concern. Lire la suite

OpenAI has acquired Odessia, an AI travel agent founded by Sonder co-founder Francis Davidson, and shut the service down the same day. Davidson announced the deal on X on Tuesday and said he will join OpenAI to focus on the future of travel. The deal coincided with OpenAI's launch of Businesses in ChatGPT, which lets companies sell through the chatbot. OpenAI confirmed the acquisition and called Davidson a proven founder. The company, less than a year old, had attracted tens of thousands of users over the summer. OpenAI has not disclosed what it plans to do with the technology or whether ChatGPT will handle travel bookings directly. Shares of several online travel companies fell after the news. Lire la suite

Ohio State University and Microsoft have signed a five-year partnership that will move up to 80% of the university’s storage and computing to Microsoft’s Azure cloud platform. The deal also expands the use of AI note-drafting tools at the Wexner Medical Center and sets up pilot “smart hospital” systems. Microsoft announced the agreement on Tuesday, but neither side disclosed financial terms. The partnership is part of a broader push by big tech companies to sell AI services to health systems, a race that intensified the same day Google Cloud launched its Gemini product for healthcare. Lire la suite

A startup lab called Physical Superintelligence (PSI) says an AI model helped solve the Aaronson-Ambainis conjecture, a quantum computing problem that had gone unresolved for 18 years. David Turturean, a recent MIT graduate on PSI's technical staff, used early access to Anthropic's Fable 5.2 model to reach the result. The proof was checked by formalizing it in Lean, a coding language for math, and reviewed by top researchers. The result rules out one route quantum computers could take to solve certain hard problems quickly, though not all of them. PSI, which currently earns revenue by using its tools to optimize data centers, says the episode shows how AI can reshape scientific research. Its leaders also acknowledge that the field has not yet settled on standards for sharing and validating AI-driven discoveries. Lire la suite

Zendesk is changing how businesses pay for its artificial intelligence software. Instead of charging for each person who uses the tools, the customer service company will charge for problems its AI resolves or tasks it completes. The change was announced Tuesday at the company's AI Summit in San Francisco, alongside a new Outcome AI Platform. Customers set an Outcome Budget and pay only for eligible results. Zendesk also introduced tools that let outside AI assistants find verified support channels and connect to company systems with human oversight, reflecting a shift in how consumers expect to handle customer service. Lire la suite

Microsoft researchers have found that fewer than 10 rarely activated neurons in a large language model can be reprogrammed into a hidden backdoor. The model behaves normally until it encounters a specific trigger, then can carry out malicious commands. The technique, called Lazarus, does not require retraining the model and can evade standard safety evaluations. The risk is greatest for AI agents with access to tools or terminals, and the researchers say organizations should verify where their models come from and check their weights against the original versions. Lire la suite

Ten artificial intelligence companies, including OpenAI, Anthropic, Microsoft and Mistral, have signed a new declaration on the safety of advanced AI models. The Paris Call for AI Safety and Security was agreed Monday at a meeting hosted by French President Emmanuel Macron on the opening day of the Paris Peace Forum. Six countries also signed, but the United States, Google, Meta and China did not. The French AI safety group CeSIA signed the declaration but says it leaves major gaps, including no common risk thresholds and no guarantee that independent testers or governments will get access to the systems they need to assess. Lire la suite

New York has opened a registration portal for the largest frontier AI developers, including OpenAI and Anthropic, as its RAISE Act moves toward enforcement. Companies must register by January 1, 2027. Under the law, developers with more than $500 million in annual revenue must publish safety frameworks, report critical safety incidents within 72 hours, and release public transparency reports. A new state office will oversee compliance, and violations can carry penalties of up to $3 million. Governor Kathy Hochul called the law a national model for AI regulation and launched a state AI research center on the same day. Lire la suite