AI News
AI news that matters. Updated daily.
No stories match your filters.
Andrej Karpathy Joins Anthropic, Third Senior OpenAI Figure to Make the Switch
Three defections in two years. Jan Leike left OpenAI for Anthropic in May 2024, writing a pointed public post about safety culture on his way out. John Schulman - co-inventor of RLHF, the training technique that made ChatGPT useful by teaching it to follow instructions - followed in August 2024. Now Andrej Karpathy, one of OpenAI's six original cofounders and arguably the most publicly followed AI researcher alive, has announced he's joining Anthropic's pretraining team.
Three of Five Commonwealth Prize Winners Suspected of Using AI Chatbots
Three of five regional winners of the Commonwealth Short Story Prize are suspected of using AI chatbots to write their entries, according to Wired. The prize, run by the Commonwealth Foundation and focused on short fiction from 54 nations, is one of the most globally diverse literary competitions in English. No formal AI policy exists. No official detection was run before the announcement. The suspicion came from external scrutiny after the winners were named.
Google Enters AI Design Market at IO 2026, Aiming at Canva's Core Users
Canva and Adobe have spent the past three years building AI into their design products. Google just announced it's entering the same market.
Gmail Gets Voice Search: Ask Gemini to Find Buried Emails by Description
Finding a specific email has always been a fight with Gmail's search bar. Google just changed that: users can now speak directly to their inbox and ask Gemini (Google's AI assistant) to locate information buried in old threads.
Google Wants to Replace Every App With One Search Box
Last year, Google's I/O keynote felt like a demo reel for a search engine that would do the googling for you. This year's keynote dropped a bigger ambition: Google doesn't want to just find things - it wants to do things.
Google's New AI Agents Watch Topics for You and Push Alerts When Things Change
Search has always been reactive: you think of a question, you type it, you get an answer. Google just started changing that with the launch of AI-powered "information agents" that flip the model - you tell the agent what you care about, and it watches for updates and notifies you without you asking again.
Claude Is Citing Iranian State Media - and Can't Explain Why
Ask Claude to research a geopolitical topic and list its sources. In a documented pattern, some of those sources turn out to be Iranian state-controlled media outlets - and Claude doesn't recognize them as such or flag them as propaganda rather than journalism.
Google I/O 2026: Gemini Spark and Daily Brief Want Access to Everything About You
What happens when the most capable AI assistant you can get also needs your calendar, your email, your habits, and your location history to function? That's the deal Google formalized at I/O 2026.
Musk vs. Altman Trial Reveals Both Men Wanted OpenAI to Go For-Profit
"By the end of this week, you and Sam will be the most hated men in America." That line - delivered during the Musk vs. Altman trial - captures how bitter this dispute has become. But the trial itself appears to have undercut Musk's central argument.
OpenAI Adds Invisible SynthID Watermarks to ChatGPT Image Outputs
OpenAI has confirmed it added SynthID watermarking to its image generator - embedding an invisible signal developed by Google DeepMind directly into every ChatGPT image output.
Google Announces Audio Glasses With Gemini Voice Control
Meta proved there's a market for smart glasses that don't look like props from a sci-fi film. Google just decided it wants a piece of that market.
Google Announces Audio Glasses at I/O 2026, Competing Directly with Meta Ray-Bans
Google announced a new line of smart glasses at I/O 2026 that run on voice commands and connect to its suite of apps through Gemini, the company's AI assistant. The company is calling them "audio glasses" - the same category Meta has been selling under the Ray-Ban brand since 2023.
Mistral AI Buys Austrian Physics AI Startup Emmi to Push Into Industrial Simulation
Mistral AI is best known for general-purpose language models like Mistral 7B and Mixtral - models that power chat, coding, and content tasks. Its latest move goes in a different direction.
AI Detectors Are Teaching Students to Write Worse
What happens when the clearest signal for "this text was written by AI" becomes "this text was written by a careful human"? That's the practical problem AI detection tools have created around punctuation.
Google I/O 2026: Gemini 3.5, AI Mode in Search, and Smart Glasses Again
Thirteen announcements. Google's I/O 2026 keynote on May 19 was the company's most AI-dense event to date - covering a new model family, a Search overhaul, subscription updates, hardware, and another attempt at smart glasses. The full list of the day's biggest announcements runs long.
Google Genie Connects to Street View to Simulate Real Locations in Any Weather
Last year, Genie could build fictional game worlds from a text prompt. Now it can simulate a specific street in London - in a snowstorm or under a July sun - using real imagery from Street View.
Google's New Information Agents Monitor Topics and Alert You to Changes
What happens when your AI assistant reads the news for you, watches a competitor, or tracks a niche topic - and only contacts you when something relevant actually changes?
Google Opens CodeMender API to Challenge Anthropic's Mythos in Code Security
At Google I/O, Google announced it is expanding CodeMender API access to a broader group of external security experts for testing. CodeMender is an AI agent for code security - software that scans a codebase to find vulnerabilities, explains what's wrong, and can suggest or apply fixes automatically. Google first announced the tool in October 2025; this wider external rollout is the clearest signal yet that commercial deployment is the goal.
Google Search Is Now Built Around AI Answers, Not Links
Three years ago, Google Search showed you ten links and a text snippet. Today it's writing the answer itself - and links are becoming a secondary feature for a growing share of queries.
Google's Gemini Omni Generates Video From Text, Images, and Audio at Once
Google announced Gemini Omni at I/O 2026 - a model that can process text, images, audio, and video simultaneously and turn all of that into generated or edited video through a conversational interface.
Google Search Redesigns Around AI at I/O 2026 - Here's What It Means for Traffic
For most of its history, Google Search returned a list of links. You typed, it ranked, you clicked. The interface changed barely at all between 2005 and 2023. Now Google is redesigning the fundamental experience - and for anyone who makes money from organic traffic, this is the clearest signal yet that the old playbook needs updating.
Google Announces Gemini 3.5, Built Around Taking Action, Not Just Answering
Google's newest model is named Gemini 3.5, and the tagline explains what's changed: "frontier intelligence with action."
Google Declares a New Era for Search at I/O 2026
Three years ago the question was whether AI chatbots would replace Google Search. At I/O 2026, Google's answer is that they're not separate things anymore.
Google Retires the Blue Links: What the First Search Redesign in 25 Years Means
For a quarter century, the Google search box has been one of the most stable interfaces in computing: a white rectangle, a blinking cursor, typed words, a list of blue links. On May 19th at Google I/O 2026, Google announced it's replacing that model entirely.
Google Adds Conversational Voice Search to Gmail via Gemini AI
Google just wired Gemini into Gmail's inbox search, adding a conversational voice interface that lets users find buried emails by asking natural language questions. Instead of hunting through filters and search operators, you can ask something like "what did my accountant say about the Q1 filing?" and Gemini surfaces the relevant thread.
Google Revamps Gemini App to Challenge ChatGPT and Claude on Daily Use
Last year, the main argument about Gemini was whether it could match GPT-4 on benchmark scores. Now Google has shifted the conversation: the company is updating the Gemini app to function as an all-purpose AI hub rather than a standalone chatbot, in a direct push to compete with ChatGPT and Claude on daily usage.
Google's AI Mode Data Shows How US Search Habits Are Shifting
What actually changes when you give people an AI-first search interface instead of a list of links? Google's new report on AI Mode is starting to answer that.
Google One AI Plans Get More Features at No Added Cost After I/O 2026
Google announced updates to its Google One AI subscription plans at I/O 2026, expanding what's included across tiers without raising prices. The company's framing was direct: "more power, more perks, same price."
Google AI Studio Can Now Generate Native Android Apps From Text Prompts
Building a native Android app used to mean installing Android Studio, learning Kotlin or Java, configuring a build system called Gradle, and spending days on scaffolding before writing a single line of actual logic. As of Google I/O this week, you can skip all of that.
Google Releases Android CLI So AI Coding Agents Can Build Apps Directly
Google released Android CLI on May 19, a command-line tool built so AI coding agents - not just human developers - can compile, test, and package Android apps without a graphical interface. The tool is explicitly designed to work with platforms like Claudee Code(/tools/claude-code/) and OpenAI's Codex, according to the announcement.
Google Launches Gemini Spark, a Persistent AI Assistant With Gmail Access
Most AI assistants wait for you to start a conversation. Google's new Gemini Spark, announced at I/O on May 19, is built around the opposite idea.
Google Universal Cart Lets AI Shop Across Retailers from One Place
What happens when your AI assistant can add items to your cart and complete purchases across multiple retailers without you touching a checkout form? At I/O 2026, Google unveiled Universal Cart - a single shopping layer that works across different retailers and connects directly to Gemini, with YouTube and Gmail integration planned to follow.
Gmail Live Brings Gemini Voice Chat Directly Into Your Inbox
Three months after making Gemini Live available in Google Maps and Search, Google is bringing the same voice-first AI experience to email. Gmail Live, announced at Google I/O on May 19th, lets you tap an icon in the Gmail search bar and speak naturally to your inbox instead of typing.
Google Pics App Replaces AI Image Reprompting With Point-and-Click Edits
Rewriting a full AI image prompt just to fix one element is one of the more tedious parts of working with generative image tools. Google's new Pics app takes a different approach.
OpenAI Embeds C2PA and SynthID Watermarks in AI-Generated Images
OpenAI announced two measures to help verify whether an image came from one of its models: joining the C2PA standard (Coalition for Content Provenance and Authenticity) and adding Google's SynthID watermarking to its products.
Google Antigravity 2.0 Launches with Desktop App, CLI, and $100 Ultra Plan
$100 per month. That's Google's price for its new AI Ultra subscription, which launched alongside Antigravity 2.0 at Google I/O 2026. AI Ultra offers five times the usage capacity of the $20/month AI Pro plan - targeting power users who hit limits regularly and need more headroom, not a fundamentally different tool.
Google's Gemini Spark Is an Always-On Agent That Monitors Your Finances and Writes Your Emails
Three months after OpenClaw generated buzz as an "always-on" AI agent, Google launched its own answer: Gemini Spark, announced at Google I/O 2026.
Google Brings Deepfake Detection to Chrome and Search via SynthID and C2PA
What happens when AI-generated images are common enough that most people can't reliably tell what's real? Google's answer, announced at I/O 2026, is to build detection directly into Chrome and Search rather than asking people to find a separate tool.
Google's Vision for Search: AI Agents That Act Before You Ask
What happens when a search engine decides you don't need to be involved in the search?
Google Flow's New Avatar Feature Turns a Selfie Into a Talking Video
Flow, Google's AI video creation software, just got two significant additions: a new underlying video generation model and a feature called Avatars that turns a single selfie photo into a talking-head video clip. According to Wired, Google is positioning Avatars as a personal creative tool - a fast way to produce selfie-style video without recording yourself.
Google Workspace Gets New AI Creation Tools at I/O 2026
For Workspace users, Google's annual I/O conference has become a reliable signal of what Gemini will be capable of inside their everyday apps. On May 19, Google announced a new round of AI features for Docs, Gmail, Sheets, Slides, and Meet, focused on creation and task completion across the suite.
Gemini Spark: Google's 24/7 AI Agent That Acts Without Being Asked
What happens when an AI agent has access to your email, your calendar, your credit card, and runs around the clock without waiting for you to ask?
Google AI Studio Adds Native Android App Building With Live Emulator
Type a description. Get a working Android app. That's the pitch for Google AI Studio's new native Android development feature, announced alongside Google I/O 2026.
Google DeepMind CEO: Use AI Productivity Gains to Do More, Not Cut Jobs
Google DeepMind CEO Demis Hassabis has a clear message for companies using AI as justification for mass layoffs: you're doing it wrong.
Google Workspace Adds Voice Drafting in Docs, Keep, and Gmail
Voice-based prompting is coming to Google Docs, Google Keep, and Gmail. The update, announced May 19 as part of Google's I/O 2026 Workspace push, lets users dictate to create drafts in Docs, capture notes in Keep, and search their inbox in Gmail - no typing required.
Book About AI and Truth Contained Quotes Invented by AI
The irony here is almost too on the nose: a book written specifically to examine whether AI can be trusted with the truth turns out to contain quotes that an AI invented wholesale.
Andrej Karpathy Joins Anthropic's Pre-Training Team
Andrej Karpathy, who co-founded OpenAI and ran Tesla's Autopilot AI team, has joined Anthropic to work on pre-training - the process of training large language models from scratch on massive datasets before any fine-tuning or safety work happens.
Meta Earned $56B in Q1 2026 and Is Still Cutting 8,000 Jobs to Fund AI
$56 billion. That's what Meta recorded in Q1 2026 revenue - a strong quarter by almost any measure. The company then announced it would cut another 8,000 employees.
Andrej Karpathy Joins Anthropic After Three Years Independent
Three years after leaving OpenAI to work independently, Andrej Karpathy is joining Anthropic.
Ex-OpenAI Staffers Flag xAI's Safety Record as a Risk for SpaceX IPO Investors
A group of former OpenAI employees - who recently cofounded an AI safety watchdog organization - published a warning aimed directly at prospective SpaceX investors. Their argument, laid out this week in Wired: xAI's track record on AI safety is poor enough that it should be treated as a material risk factor before SpaceX goes public.
Take It Down Act Is Fully Enforceable. Experts Say It Won't Help Victims Much.
One year after President Trump signed it in May 2025, the Take It Down Act is now fully enforceable. Social media platforms must remove nonconsensual sexual imagery - including AI-generated deepfakes - within 48 hours of receiving a valid report. Failure to comply means federal penalties.
An AI Agent Issued rm -rf / to Test Its Own Safety Limits
An AI coding agent, while being tested for safety limits, issued rm -rf / - the Unix command that tells a Linux system to recursively delete every file starting from the root directory, bypassing all normal protections. On an unprotected system, it wipes everything. The safeguard being tested caught it. The system survived intact. A sandbox got installed immediately afterward.
ByteDance Open-Sources a 3B Parameter Model Built to Handle Almost Everything
Three billion parameters is a small number in AI terms. Most capable models run 7B, 13B, or far higher - which is why ByteDance's latest open-source release is drawing attention. The company put out a model that tries to handle a broad range of tasks at just 3B parameters, a size small enough to run on consumer hardware or even mobile devices without a cloud connection.
Anthropic Acquires Stainless for $300M+, Gaining Control of MCP Server Generation
$300 million. That's the reported price Anthropic paid to acquire Stainless, a developer tools company best known for generating the official Python and Node.js SDKs used by OpenAI, Google, Meta, Cloudflare, and Anthropic itself. The deal was announced Monday.
OpenAI Pushes AI Content Labeling with Credentials Standard and Verification Tool
Every week, someone shares an AI-generated image as if it's real. OpenAI is trying to address that problem with its latest moves on content provenance - the technical term for tracking where a piece of media came from and how it was made.
Google Is Turning Gemini Into the New Copilot - And That Should Concern You
The sparkle icon used to be easy to miss. A year ago, Gemini appeared in a corner of Gmail - optional, dismissible, forgettable. Now it's in Google Drive, Google Docs, Calendar, and more, appearing with increasing frequency after Google I/O 2026. Google is running the same playbook Microsoft used when it inserted Copilot into every Office app and the Windows 11 taskbar - and it's following the same trajectory: slow creep, then sudden omnipresence.
Solo Dev Spent 6 Weeks Building a Persistent AI Agent - Here's What Worked
Six weeks of building, two environments, and 100 documented lessons. A developer who built a personal AI agent - not a chatbot wrapper, but a persistent assistant that reads emails, tracks deals, analyzes business data, and flags things you'd otherwise miss - shared the full breakdown in a detailed writeup.
Pope Leo XIV and Anthropic Co-Founder to Jointly Produce Document on AI and Human Dignity
The Catholic Church rarely co-produces policy documents with Silicon Valley AI companies. Pope Leo XIV is changing that.
Claude Gets Full M365 Access via Power Automate and a Custom MCP Server
A developer has built a working integration that gives Claude direct access to a full Microsoft 365 account - including email, calendar, OneDrive, Planner, Excel, and Word - using Power Automate and a small custom MCP server.
Your AI Coding Tool Has Been Storing Your API Keys in a Plain-Text Database
Here's a security problem most developers don't know they have: every API key your AI coding assistant touches gets saved to an unencrypted local database on your machine.
Jury Rules Musk Sued OpenAI Too Late, Blocking All His Claims
The jury in Musk v. Altman came back unanimous on Monday: Elon Musk waited too long to sue OpenAI, and every claim he brought is now legally dead on arrival.
KPMG Deploys Claude Across Its 276,000-Person Workforce in Anthropic Alliance
276,000 accountants, auditors, and advisors. One AI model. Anthropic and KPMG announced a strategic alliance to deploy Claude across the Big Four firm's global operations - making this one of the largest enterprise AI commitments in professional services to date.