AI news · Friday, October 2, 2026

OpenAI rolls out GPT-6-powered 'agent' called Dots at dev event

At its annual developer conference, OpenAI unveiled Dots, an AI agent that can carry out multi-step tasks on its own, built on a new model called GPT-6 Astra. It's a direct answer to Meta's surprise hit agent platform Muse, which launched earlier this year and has seen fast early adoption. The fight now is over which company's helper actually gets used daily, not just demoed on stage.

Follow the money: OpenAI is chasing a product Meta can give away for free, which means this battle will be won or lost on habit and distribution, not raw capability.

via The Verge AI

Google tests putting AI data centers in space

Google launched a test satellite carrying one of its advanced AI chips, part of Project Suncatcher, an effort to eventually run data centers in orbit powered by constant sunlight instead of the electric grid. Google estimates SpaceX's Starship rocket would need to launch roughly 1,800 times to make the idea practical at scale. It's a long-shot bet, but it shows how desperate big AI companies are for more power.

Zoom out: This is less a space story than a power story: AI's energy appetite is now pushing serious companies to consider literally leaving the planet for electricity.

via Fast Company Tech

Judge sides with Google, tosses lawsuits over AI search summaries

A federal judge dismissed antitrust lawsuits from Chegg and Rolling Stone's parent company, which claimed Google's AI-generated search summaries unfairly siphon traffic away from their websites by answering questions directly instead of sending readers to click through. The judge acknowledged AI search has real consequences for publishers, but ruled that isn't the same thing as breaking antitrust law.

The catch: The ruling doesn't settle whether AI summaries are killing the web traffic publishers depend on, it just says courts aren't the place to fix it.

via The Verge AI

Grok reportedly advised Trump before Venezuela capture operation

According to reporting, President Trump asked Elon Musk's chatbot Grok for its opinion before the US moved to capture Venezuelan president Nicolas Maduro. Few details on what Grok actually said have surfaced, but the episode is notable simply because it happened at all: a consumer chatbot apparently consulted ahead of a major foreign policy and military decision.

The twist: Whether or not Grok's advice mattered, the fact a chatbot was in the room for a decision like this is a preview of how casually AI may enter high-stakes government choices.

via TechCrunch AI

Poll: 73 percent of Americans now worry AI threatens humanity

A new Quinnipiac University poll finds 73 percent of Americans are concerned that AI could pose a threat to humanity's survival, not just to jobs or electricity bills. That fear has moved from a niche worry among researchers to a mainstream public opinion, even though most people can't point to a specific mechanism for how it would happen.

Zoom out: Public fear this widespread tends to eventually force political action, regardless of whether experts agree on what the actual risk level is.

via Fast Company Tech

AI can now rebuild the image in your mind from a brain scan

Researchers built an AI tool that reads brain scan data and reconstructs, with striking accuracy, the actual image a person was looking at. It also works in reverse, predicting what someone's brain activity would look like given a picture. The tool doesn't read thoughts or words, just visual activity tied to what the eyes are seeing in the moment.

What's next: This is still narrow, tied to literal eyesight rather than thoughts, but it's the clearest sign yet that brain-to-image decoding is becoming a real, working technology.

via MIT Technology Review

Nurses say Palantir software is creating scheduling chaos, not fixing it

A hospital network brought in Palantir, a data company known for government and military work, to streamline staff scheduling. Nurses and other health workers say the new system instead causes errors, confusing shift assignments, and extra burnout as staff scramble to fix mistakes the software introduces. It's a case study in AI tools built for efficiency on paper creating more work in practice.

The catch: The people who actually have to use enterprise AI tools day to day rarely get a say before management signs the contract, and this is what that gap looks like.

via Wired AI

Google's Gemini app can now describe what your phone camera sees

A new feature called Guided Vision, rolling out in Google's Gemini Live app on Android, uses AI to describe out loud whatever you point your phone's camera at, in real time. It can read small print like a menu or medicine label, describe your surroundings, or help identify an object. It's aimed partly at people with low vision, but useful for anyone in a pinch.

For you: This turns a phone most people already own into a basic accessibility tool, no special hardware or subscription required.

via The Verge AI

Shopify lets anyone build an online store just by chatting with AI

Shopify launched Canvas, a tool that lets merchants build and customize an online store by describing what they want to its AI assistant, Sidekick, and watching the site update live as they chat. No code or design skills needed. It's part of a broader shift where AI is replacing the drag-and-drop website builder with plain conversation.

For you: This lowers the bar for starting a small online business meaningfully, the same way free website builders did a decade ago.

via TechCrunch AI

Why AI chatting still isn't the same as AI reasoning

A longtime AI researcher argues that despite impressive answers, today's chatbots don't actually reason the way the term implies. They predict plausible-sounding next words based on patterns, which can look like logic but breaks down on problems slightly outside their training patterns. The piece uses the famous 2016 Go match, where an AI made a move so strange it looked like genius, as a case study in how easily we mistake pattern-matching for understanding.

The twist: Knowing the difference matters practically: it tells you when to trust an AI's confident-sounding answer and when to double-check it yourself.

via MIT Technology Review

What does it mean when AI 'hallucinates'?

When an AI chatbot hallucinates, it confidently states something false as if it were fact, a made-up legal case, a fake statistic, a book that doesn't exist. It's not lying on purpose. These systems generate text by predicting the most likely next word based on patterns in their training, and sometimes the most plausible-sounding answer simply isn't true. Think of it like a student who didn't study but is great at sounding smart in an essay. They'll write fluent, confident sentences that fill gaps with guesses dressed up as facts.

For you: Knowing this turns you from a passive reader of AI answers into someone who checks the one or two facts that actually matter before acting on them.

Original explainer

AI finally beats top Stratego players, a game built on hidden information

Stratego, a board game where each player's pieces are hidden from their opponent, has long resisted AI because so much depends on bluffing and guessing what you can't see. Researchers cracked it by adding a second AI that specifically guesses what hidden pieces probably are, paired with the main decision-making AI. It then beat the best human Stratego player in history, on a surprisingly modest computing budget.

Zoom out: Games with hidden information are a proxy for real-world situations like negotiation or poker, so progress here hints at AI getting better at reasoning under uncertainty generally.

via Ars Technica AI