Actualités — Page15

Anthropic's 335-page S-1 prospectus, filed Monday, pairs sweeping promises about advanced AI with some of the most sobering risk disclosures Wall Street has seen. The founders urge investors to see their stock purchase as "an opportunity to be part of something greater," while the filing warns that AI models could pose "catastrophic or existential risks to humanity." The document includes 94 pages of risk factors, and it omits some key financial details, including third-quarter results. Analysts who have studied past tech IPOs say the length and tone of the filing stand out, even as the company's survival depends on raising large sums of money. Lire la suite

Sens. Ron Wyden and Edward Markey are investigating what they call "serious privacy and safety risks" tied to Meta's AI smart glasses. In a letter to CEO Mark Zuckerberg, the senators cite reports of men secretly filming women with the glasses and of Meta's AI contractors reviewing footage showing people undressing and having sex. They say Meta launched the product even after identifying risks to bystander privacy, and they have asked the company to answer a series of detailed questions by Thanksgiving, November 26, 2026. Lire la suite

Starbucks is exploring technology that could recognize customers as they pull up to the drive-thru and display a personalized digital menu. CEO Brian Niccol outlined the plan on Thursday during an earnings call, describing “experiences that recognize participating customers as they arrive, present personalized digital menus, and make payment more effortless.” The company has not said when the feature might reach stores. Niccol described the effort as still “early,” adding that Starbucks will “test, learn, and scale what works.” The move follows similar drive-thru AI experiments by Taco Bell and McDonald’s. Lire la suite

Google's prototype space data center, Project Suncatcher, is online and communicating with ground teams after launching earlier this month on a SpaceX Falcon 9 rocket. Engineers sent the satellite's pre-loaded Gemma 4 model a request to write a haiku about being in space, and it responded. The satellite carries four Google Tensor Processing Units, and the next tests will examine how the chips handle heat and radiation in orbit. The company plans to launch two more satellites in 2027 to test interconnected operations, a step toward putting AI computing in space. Lire la suite

The Wikimedia Foundation, which runs Wikipedia, is asking a federal judge to separate two questions in The New York Times’ copyright fight with OpenAI. The foundation says training AI models on copyrighted work is likely fair use, but a chatbot that copies websites to answer user questions likely is not. In a friend-of-the-court brief, Wikimedia takes neither side. It warns that broad rulings for either publishers or AI companies could threaten free knowledge and the open web. The judge has not yet ruled on the fair use question or on whether to accept the brief. Lire la suite

New York has banned AI chatbots from pretending to be human when talking with children, and from building emotional dependency, encouraging self-harm or engaging in sexual conversations with them. Governor Kathy Hochul signed the law on Wednesday, alongside a second law requiring security features on internet-connected baby monitors. The chatbot restrictions apply to users under 18 and to anyone whose age has not been verified. State Attorney General Letitia James can seek civil penalties of up to $25,000 per violation. The law takes effect on January 1, and tech industry groups have pushed back on how broadly it is written. Lire la suite

The Biological Computing Co., a San Francisco startup known as TBC, has released TBC Mischief 1, an AI model that turns still images into short videos. The company says lessons from lab-grown neurons make the model about five times faster than the open video model it is built on, Wan 2.2, at the same quality. TBC prices the model at $0.005 per second of video and says it has not published independent tests to verify the speed claim. The launch follows a September deal with Amazon Web Services to sell its models to businesses. Lire la suite

Pope Leo XIV says artificial intelligence systems should be “disarmed” and that humans have a responsibility to reject AI applications when necessary. In a papal address delivered Friday, he urged United Nations member states to take concrete steps toward banning lethal autonomous weapons ahead of their meeting in Geneva next month. The pope said decisions involving human life, freedom and the use of force must remain under human oversight. He also warned against granting AI systems rights or legal personhood, arguing that machines are tools, not persons. Lire la suite

VRChat will not add labels or filters for AI-generated content, even though 95 percent of users who took a company survey said they want some form of AI label. In a support post, the virtual world platform said it will keep weighing a "Human-authored" label and a "Tools used" field but has not settled on either. The company said labels could lead to users hunting for AI content and harassing creators, even those who use no AI tools. Instead, it plans changes focused on accurate thumbnails, better discovery tools and quality. Lire la suite

Amazon CEO Andy Jassy said Thursday that Amazon Web Services can grow into a trillion-dollar revenue business within five to 10 years. He made the case during the company's quarterly earnings call, where analysts pressed him on the spending needed to get there. Amazon expects about $220 billion in cash capital expenditures in 2026, most of it for AWS, and plans to spend meaningfully more in 2027. AWS revenue rose 39% from a year earlier. Investors were unconvinced, and Amazon shares fell about 6% in after-hours trading. Lire la suite

Reddit CEO Steve Huffman says Google’s AI Mode and AI Overviews are cutting the search traffic that once flowed to the platform, with referrals from search falling by more than 1 million a week in October. In response, Reddit will begin testing a logged-in-only experience outside the United States, a move Huffman says will improve user experience, ad targeting and protection of the company’s data. The change marks the company’s most dramatic shift yet as it contends with AI scraping and changes to internet search. Lire la suite

Baseten has released joey-1, a small open-weights AI model that answers yes-or-no and multiple-choice questions instead of writing text. The company says the 2-billion-parameter model costs $0.01 per million tokens and responds to a typical 2,000-token request in about 20 milliseconds. Baseten is pitching it for cheap, fast jobs such as routing support tickets, spotting spam and blocking harmful prompts. It joins two larger models, mammoth-1 and elephant-1, released last week by the company's research lab, Base Labs. Lire la suite

Meta has quietly launched Style Sketch, an AI fashion app that suggests outfit combinations using clothes users already own. The app can pull a virtual wardrobe from a user's Instagram photos or from items added by camera or camera roll, and it can preview looks on a model matched to the user's body shape and skin tone. It is available in Indonesia, Singapore and Thailand on iOS and Android. Meta has not said whether it plans a wider release. The launch is one of several recent AI apps from the company and fits a broader push to ship more apps faster. Lire la suite

Samsung Electronics reported a record operating profit of 107.4 trillion won, about $80 billion, for the third quarter, as demand for memory chips used in artificial intelligence data centers kept climbing. Revenue also hit a new high of 195.3 trillion won. Nearly all of the profit came from the chip business, while the company's mobile, television and home appliance units posted losses. Samsung said it will keep spending heavily on new plants and equipment and declared a quarterly dividend of 4,601 won per share. Rival SK Hynix also reported record profit this week. Lire la suite

The US power sector could release 42 percent more planet-heating carbon dioxide in 2030 than it did in 2025, as utilities build more fossil fuel infrastructure to meet data center electricity demand, according to a new analysis by the Sierra Club. The group says the increase would wipe out roughly 15 years of greenhouse gas reductions. The estimate is higher than some earlier projections, and the final outcome depends on how many planned data centers are built and what technology powers them. Lire la suite

French artificial intelligence company Mistral says its model Le Chonk is now the best open-weight model in the world for automated coding and cybersecurity. The company says the model matches or outperforms the top offerings from DeepSeek and Z.ai in those areas. Mistral first described the model as the most advanced open-weight offering outside China when it released an early version a few weeks earlier. The new claim follows further refinement of the model. Lire la suite

A test of five major AI assistants found that most of the statistics they cite cannot be traced to the research behind them. Trakkr Research asked ChatGPT, Perplexity, Gemini, Claude and Google's AI Overviews 200 questions that call for a number in October 2026, then followed the citations for 1,285 statistics. Only 40% led to a study that said the same thing. Many of the rest ended at pages that repeated the figure without citing any research, and some of the most widely repeated numbers had no study behind them at all. ChatGPT performed best, with 57% of its statistics sourced. The findings suggest that a confident answer and a citation do not guarantee that a number is reliable. Lire la suite

Ottawa startup Solink has raised more than $100 million to expand software that turns the security cameras already installed in shops, restaurants and warehouses into business data. Vertu Capital led the round, with Goldman Sachs Alternatives, Golub Growth and Comcast Ventures joining. The company says its AI platform now runs in more than 35,000 locations across 65 countries. Solink plans to spend the money on new AI features and what it calls strategic opportunities, as rivals such as Coram and Veesion also attract investment for camera-based AI. Lire la suite

Death Clock, an AI-powered app, promises to predict when users will die and help them extend their time on Earth. WIRED senior writer Jason Parham tried it, giving blood in the process. He was told he is scheduled to die on August 10, 2060, at age 74. After a month with the app, Parham says it felt like it was sucking the life out of him. Lire la suite