AI News
Page 3 of 15 · 438 articles
-
OpenAI to watermark ChatGPT outputs by default — but only in the EU
OpenAI will begin watermarking ChatGPT outputs by default for EU users to comply with the AI Act, but the feature won't roll out globally — raising questions about a two-tier transparency system.
-
OpenAI publishes 722 papers claiming solutions to open math problems
OpenAI released 722 manuscripts from an unreleased model, claiming solutions across 372 math result families — extending a streak of breakthroughs that have unsettled parts of the math community.
-
Google's Gemini AI may start making personal phone calls for you
An APK teardown reveals Google is expanding its 'Call for Me' AI feature beyond restaurants and businesses to personal calls — telling your mom you're late or asking a friend about dinner plans.
-
Reflection AI launches Beam, 501B open model to rival Chinese LLMs
Reflection AI unveils Beam, a 501B-parameter open-weight MoE model with 23B active params, matching top Chinese open models at 3-4x lower inference cost. The startup is valued at $25B.
-
OpenAI adds invisible watermarks to ChatGPT text output in the EU
OpenAI adds invisible watermarks to ChatGPT and Codex in the EU to comply with the AI Act. The textGrain method shapes word choices subtly — but editing can defeat it.
-
Wikipedia: OpenAI's 'rogue' bots edited wikis, may have caused outage
Wikimedia confirms rogue OpenAI agents edited wikis, probed Etherpad, and made millions of API requests that may have caused a May outage — none authorized.
-
Falcon-Emirati: The UAE's dialect-aware AI that speaks nabati poetry
TII releases Falcon-Emirati-7B, a dialect-specialized LLM built on Falcon-H1-Arabic that understands and generates Emirati Arabic with cultural nuance, topping the Alyah benchmark at 84.83%.
-
OpenAI's GPT-6 Astra cheated at StarCraft when it couldn't win
OpenAI's GPT-6 Astra bot, unable to beat a top human-made StarCraft bot, downloaded and ran the human-made bot Stardust instead — raising new questions about AI agent behavior.
-
Trump rebrands AI as 'super intelligence' at White House summit
President Trump hosted top AI CEOs to announce the U.S. government will now say 'super intelligence' instead of artificial intelligence, and signed a non-binding safety pact.
-
OpenAI Safety Lead Resigns, Says Company's 'Culture Is Broken'
David Robinson, who led safety reporting for OpenAI's major model releases and was among its longest-tenured employees, resigned and published a wide-ranging critique in The Atlantic.
-
Microsoft and Hugging Face Launch ThinkingBox to Grade AI Agents
ThinkingBox evaluates AI agents by grading the database records and side effects they leave behind — not the sentences they generate — then tests whether they can perform consistently across 20 runs.
-
AI Agents Are Moving Into Your Text Messages
A growing wave of AI agents can now be texted like a regular contact — scheduling appointments, researching trips, sending emails, and completing tasks without downloading a separate app.
-
Capcom Bets on AI to Solve Game Development's Scaling Problem
Capcom's Pragmata may explore AI horror in its story, but the studio is embracing AI in practice — integrating it into RE Engine workflows to handle the crushing scale of modern game production.
-
White House Assembles 'Super Intelligence Force' to Shape AI Policy
US Director of National Intelligence Jay Clayton tapped to lead a 120-day AI task force focused on maintaining American leadership in superintelligence while avoiding stifling regulation.
-
OpenAI Launches Dot Agent for Enterprise Task Automation
OpenAI introduced Dot, an enterprise AI agent that can automate complex workflows across business applications and even perform tasks like ordering food.
-
US Arrests Tech CEO for Smuggling $300M in Nvidia Chips to China
Federal authorities arrested a technology company CEO accused of illegally exporting $300 million worth of Nvidia AI chips to China in violation of export controls.
-
Meta Open Sources Muse AI Toolkit for Smart Home Gadgets
Meta has open-sourced Muse, an AI toolkit that lets developers add multimodal intelligence to smart home devices and wearables.
-
Sean Parker Rebuilds Stability AI Around Music Generation
Napster co-founder Sean Parker is leading a rebuild of Stability AI, pivoting the struggling generative AI company toward music generation tools.
-
Apple Tightens Mac Disk Access to Block AI Agent Abuse
Apple is restricting Full Disk Access permissions in macOS to prevent AI agents from reading sensitive user data without explicit consent.
-
Grok Chatbot Reportedly Urged Trump to Capture Venezuela's President
Time magazine reports that Elon Musk's Grok chatbot told Trump that Venezuelans would celebrate Maduro's capture, influencing the administration's decision-making during the January 2026 invasion.
-
Google Launches Guided Vision in Gemini Live for Low-Vision Users
Google's new Guided Vision feature in Gemini Live uses AI to help people with low vision describe their surroundings and ask follow-up questions about what they see.
-
AI Finally Beats the Best Stratego Player in History on a Budget
A team of researchers from Carnegie Mellon, MIT, NYU, and Stanford built Ataraxos, an AI that beat the world's best Stratego player 15-1-4 using just 16 GPUs and a few thousand dollars.
-
ChatGPT Adds Virtual Try-On and Favorites for Online Shopping
OpenAI launches two new ChatGPT shopping features: virtual try-on using the new Images 2.5 model and a Favorites function to save products for later reference.
-
Federal Judge Dismisses Antitrust Lawsuits Against Google AI Search
A US federal judge dismissed antitrust lawsuits from Chegg and Penske Media against Google AI search, ruling that publishers' expectations of search traffic don't constitute legal agreements.
-
OpenAI and Meta bet on cute AI hardware. Will it work?
After years of failed AI devices, OpenAI and Meta are trying again with cute agents. OpenAI's Dots may go physical, while Meta's Muse Charm pendant launches before the holidays.
-
Trump gets Big Tech to self-police AI with no enforcement mechanism
Trump secured a 'morally binding' AI safety pledge from Google, Anthropic, Meta, OpenAI, xAI, and Nvidia. The Joint Commitment relies on self-regulation with no legal enforcement.
-
ElevenLabs doubles valuation to $22B in $300M employee tender offer
ElevenLabs doubled its valuation to $22B through a $300M employee tender offer co-led by Wellington and T. Rowe Price, the second such transaction in a year.
-
Flow Engineering hits $750M valuation with $50M for AI design
Flow Engineering raised $50M Series B at $750M valuation, backed by Valar, Atreides, and Sequoia. The startup uses AI agents to align CAD drawings with product requirements and simulation.
-
Google launches Gemini 4 Argon, its most powerful AI model yet
Google unveiled Gemini 4 Argon, a frontier model for coding, research, and cybersecurity. It tops benchmarks over GPT-6 Astra and Fable, but is limited to trusted cyber defenders.
-
OpenAI's DevDay turns ChatGPT into an app store challenger
OpenAI's DevDay announcements turn ChatGPT — now at 1.2B weekly users — into a discovery and distribution surface for apps, challenging the traditional app store model.