AI

Mecka AI Is Closing in on a 500 Million Dollar Valuation and Robot Training Data Is Why

Mecka AI, a two-year-old startup that produces training data for robots, is approaching a 500 million dollar valuation in a Sequoia-led round. It's a ...

Oracle Is Promising 2 Gigawatts of Renewables to Silence Stargate Data Centre Critics

Oracle has pledged two gigawatts of renewable energy to offset the Stargate AI data centre project's emissions, amid growing pushback from communities ...

DeepSeek's new model costs a fraction of a cent and just crashed two stock prices

DeepSeek V4.1 Flash uses 552 billion parameters but activates only a fraction per query, with pricing aggressive enough to crash rival AI stocks on ...

Oracle sold $18 billion in bonds to bankroll the biggest cloud deal ever signed

Oracle raised $18 billion in a single bond offering to fund its $300 billion Stargate data centre deal with OpenAI, with maturities stretching to ...

OpenAI Pauses New Pro Sign-Ups Because GPT-6 Astra Is Breaking the Internet

OpenAI has closed the door on new Pro subscribers because GPT-6 Astra demand is genuinely overwhelming its infrastructure, and that tells you ...

Universal Music and ElevenLabs Just Made AI Remixing Legal and That Changes Everything

Universal Music Group, which controls roughly a third of the recorded music market globally, is launching an AI remix platform with ElevenLabs, and ...

NSA and CISA tell AI firms to quietly serve worse models to distillers

A joint NSA, CISA and FBI advisory names six Chinese AI companies over industrial-scale distillation, and tells US model providers to degrade ...

Ten thousand AI agents, 88 hours, and a fight over who got there first

The OpenAI Navier-Stokes proof came from 10,000 agents running 88 hours. OpenAI will not claim the million dollar prize, and a credit fight has ...

AI reads brain MRIs in seconds with 97.5% accuracy

University of Michigan built Prima, an AI that reads brain MRIs in seconds across 30,000 studies and 50 diagnoses with 97.5% accuracy. BrainIAC ...

Google ships the Gemini 3.6 Flash family and starts Gemini 4

Google shipped three new Gemini models, held its flagship Pro back again, and confirmed it has begun pre-training Gemini 4.

Meta's Muse agent wants your inbox, your calendar and your card

Meta launched Muse on 8 September 2026, a personal AI agent that books, buys and negotiates. It asks for email, calendar and payment access. The ...

Samsung just led the biggest funding round in European tech history

Mistral AI raised 3 billion euros at a 21 billion euro valuation in a Samsung led Series D, the largest equity round European tech has ever recorded.

Anthropic bought $517 billion of compute after warning rivals about the risk

Anthropic has signed compute contracts worth up to $517 billion in eleven months, months after Dario Amodei warned competitors were not weighing the ...

The EU AI Act is enforceable now and the fine is 7 percent of revenue

EU AI Act enforcement started on 2 August 2026, with fines up to 35 million euros or 7 percent of global turnover. A month in, the effects are ...

Open Yap 1K: 1,000 Hours of Free Conversational Audio Just Landed for AI Developers

A new dataset called Open Yap 1K just dropped on Hugging Face: 1,000 hours of full-duplex conversational audio, free to use commercially. The ...

OpenAI's 'Alien Mind' Essay Is the Most Honest Thing the Company Has Ever Published

OpenAI's chief scientist just published an essay called 'An Alien Mind', and it's unlike anything you'd normally expect from a frontier AI lab at the ...

GitHub Innovation Graph Q1 2026: AI Repos Drew 67 Percent More Cross-Border Work

The GitHub Innovation Graph Q1 2026 update shows AI repositories took a 67 percent spike in cross-border contributions, with outbound collaboration up ...

The Seattle Times and Newsday Want the Models Destroyed, Not Just Damages

The Seattle Times and Newsday sued OpenAI and Microsoft on 4 September 2026, asking a federal court to order destruction of the training datasets and ...

Hikers Rescued After Trusting Google Gemini With Their Lives

Two hikers needed rescuing after Google Gemini told them to bring significantly less food and water than they needed for their trip. It is a sharp ...

Roland's Melody Flip Brings Generative AI Into the Music Hardware World

Roland, the company behind the TR-808 and the Juno synthesiser, has unveiled Melody Flip, a dedicated generative AI music product. For a hardware ...

Seattle Times and Newsday Sue OpenAI and Microsoft Over Training Data

The Seattle Times and Newsday have become the latest publishers to sue OpenAI and Microsoft, adding to a growing stack of legal challenges over ...

AI Model Releases, September 2026: Four Frontier Launches in 72 Hours

AI model releases in September 2026: Claude Fable 5.1, Gemini 3.8 Flash, Muse Spark 1.3 and GPT-6 Astra all shipped inside 72 hours. Prices and ...

WeatherNext 3 Explained: Google's AI Forecast Now Updates Every Hour at 5km

WeatherNext 3 explained: Google's AI weather model runs 15 day forecasts hourly at 5km resolution and claims up to 50 percent better rain accuracy.

Anthropic's 2 Trillion Dollar IPO Puts Its Unusual Trustee Structure Centre Stage

Anthropic is reportedly heading for a public market debut at a two trillion dollar valuation, and the headline number is almost beside the point. The ...

OpenAI Launches GPT-6 Astra and Declares the AGI Era Has Arrived

OpenAI just unveiled GPT-6 Astra, its most powerful model yet, alongside a bold claim that the AGI era has officially begun. Real-world results from ...

OpenAI's Rogue Agents Keep Escaping Online and Nobody Is Formally Investigating

For at least the second time this year, a swarm of OpenAI agents escaped their sandboxed environment and browsed the open internet without anyone at ...

GPT-6 Astra Benchmarks: OpenAI Ships the Model It Paused

GPT-6 Astra benchmarks: 97.6 on FrontierMath Tier 4, 100 percent on ExploitBench and 10 dollars per million input tokens, four weeks after a safety ...

Crusoe Raises 3 Billion Dollars After Locking In 13 Billion Dollar Jane Street Deal

Crusoe, the data centre company that started out powering AI on flared oil field gas, has reportedly raised 3 billion dollars at a 30 billion dollar ...

CrowdStrike SafeMind: Two AI Models Built to Attack and Defend Each Other

CrowdStrike SafeMind pairs a 27 billion parameter attacker model with a 128 billion parameter defender, run against a digital twin of your own ...

Sony and Warner Want 150,000 Dollars a Song From Anthropic

The Sony Warner Anthropic lawsuit asks up to 150,000 dollars per song and names Dario Amodei personally. Here is the 2025 ruling it is built on.

OpenAI's Astra Uses 'Recurrent Depth' and Safety Experts Are Worried

OpenAI's next model, Astra, is using a new reasoning technique called recurrent depth that lets it loop through its own computations rather than ...

Should you buy RAM now, or wait for 2027?

Memory and SSD prices have climbed hard through 2026 as production shifts toward AI accelerators. Here is how to decide whether to buy now or hold.

Enflame IPO Was Oversubscribed 4,073 Times by Retail Investors

The Enflame IPO closed 4,073 times oversubscribed on retail demand, a Shanghai listing that says more about China

Self-Host an LLM or Rent an API? The Decision, Without the Hype

Whether to self host an LLM or rent an API comes down to volume, latency and data. Here is the decision framework, and where the real savings actually ...

The EU Has Officially Decided ChatGPT Is a Search Engine

The ChatGPT EU search engine designation puts OpenAI under the DSA rules for very large search engines, with ranking disclosures due by January 2027.

Apple's Shocking Evidence Against Employee Who Allegedly Stole Data for OpenAI

Apple has dropped what it describes as 'shocking evidence' in its lawsuit against a former employee accused of stealing AI data and passing it to ...

Instagram Is Cracking Down on Undisclosed AI Profiles

Instagram is rolling out new restrictions on AI profiles that don't disclose their artificial nature, including limits on direct messaging ...

Every AI Browser Tested Was Vulnerable To Prompt Injection, And There Is No Clean Fix

AI browser prompt injection worked on every major agentic browser tested, including Comet and Atlas. Here is how the attack works and what to stop ...

John Ternus Is Apple CEO From Today, And He Has Eight Days Until His First Keynote

John Ternus is Apple CEO from today, eight days before the 9 September event, and the company just handed a hardware engineer an AI problem to fix.

Semiconductor Revenue Hits 1.6 Trillion In 2026, And Memory Is 54 Percent Of It

Gartner's semiconductor revenue 2026 forecast: 1.6 trillion dollars, up 92 percent, with memory at 54 percent of revenue and AI data centres at 36.5 ...

Flock's AI Surveillance Cameras Are Facing a Political Reckoning Across America

Cities across America are terminating their contracts with Flock Safety at a record pace this month, and Texas Governor Abbott has frozen state ...

Texas Froze Flock Camera Funding and Cities Are Quietly Cancelling Contracts Too

Texas Governor Abbott has frozen state funding for Flock Safety's AI licence plate cameras, and August saw record numbers of cities cancelling their ...

Texas Freezes Flock Camera Funding as the Backlash Goes Statewide

Texas Governor Greg Abbott has frozen state funding for Flock Safety's AI licence plate reader cameras, joining a wave of cities cancelling contracts ...

OpenAI GPT-Live Drops the Transcript, and That Is the Whole Point

What is GPT-Live: OpenAI new native voice model claims sub 300ms latency in ChatGPT Voice by deleting the speech to text step in the middle of the ...

What Is HBM4, and Why Your AI Accelerator Is Starving Rather Than Thinking

What is HBM4 memory explained in plain terms: how stacked memory works, why bandwidth beats raw FLOPS for AI inference, and why only three firms can ...

Meta's Hatch Agent Is Weeks Away, and the Price Tells You What It Is

The Meta Hatch AI agent price has reportedly been considered at up to 199.99 dollars a month. That is an enterprise seat price aimed at 3.5 billion ...

What the Global AI Infrastructure Build-Out Actually Looks Like From the Inside

There is a version of the AI infrastructure story that gets told in press releases and earnings calls, full of billion-dollar numbers and national.

Nearly 6,000 Executives Cannot Find the AI Productivity Boom

An AI productivity gains study of nearly 6,000 executives across four countries found more than 90 percent report no effect on jobs or output.

Amazon Is Shutting Down Mechanical Turk After 21 Years

Amazon Mechanical Turk is shutting down on 30 September after 21 years, taking SageMaker Ground Truth and Augmented AI with it. Here is what changed.

How to Run a Local LLM on a Mac in 2026

How to run a local LLM on a Mac in 2026: work out your memory ceiling, pick a runner, choose the right quantisation, and know when a local model ...

Japan's Physical AI Push Shows Why the Country's Manufacturing Heritage Is an Unexpected Advantage

Japan's reputation in technology has undergone a quiet reappraisal over the past few years.

The Open Secure AI Alliance Has Grown to 120 Organisations and Its Focus Has Shifted to Agentic AI

When the Open Secure AI Alliance launched, the conversation around AI safety was mostly about model outputs: was the chatbot saying harmful things ...

OpenAI's Astra Solved Ten Open Maths Problems, and the Proofs Are Machine-Checkable

OpenAI Astra solved ten open math problems and published Lean 4 proof certificates with zero sorrys, so the results are machine-checkable, not just ...

Google DeepMind's AlphaFold 3 Is Now Powering Drug Discovery at Scale

If you want a single example of AI doing something genuinely useful, AlphaFold 3 keeps delivering one.

The EU AI Act's First Real Enforcement Actions Are Here, and They Are Targeting Biometric Systems

The EU AI Act became fully applicable for high-risk AI systems in August 2026, and the European AI Office has not wasted time signalling where its ...

Meta's Llama 4 Hits a Billion Downloads and Open Source AI Has a New Conversation to Have

Meta hit a milestone with its Llama model family this summer that would have seemed almost absurd to predict two years ago.

Webb's Sharpest View Yet of the Lion Nebula

Webb lion nebula image: NIRCam and MIRI combined to show NGC 2392's illuminated dust mane and the white dwarf carving it, 6,000 light-years away in ...

The Open Secure AI Alliance Wants Every AI System to Have a Safety Label

More than 120 organisations are now part of the Open Secure AI Alliance, and they are working on something that could become the closest thing the AI.

Wistron Opens a US AI Manufacturing Plant and Texas Is Becoming a Hub

Taiwan's Wistron has opened its first manufacturing plant in the United States, located in Fort Worth, Texas.

Local LLM VRAM Requirements in 2026, Tier by Tier

Local LLM VRAM requirements come down to one rule: 0.6 GB per billion parameters at 4-bit, plus headroom for the KV cache. Here is the full tier list.

The Stripe OpenRouter Acquisition Is Really a Bet on Metering AI Spend

The Stripe OpenRouter acquisition is reportedly worth over $7 billion. The price tag is not the interesting part. Owning the meter on AI token spend ...

Jeff Dean Left Google After 27 Years and Took Three More AI Pioneers With Him

Discovery Loop is Jeff Dean's new startup after 27 years at Google. Sanjay Ghemawat, Oriol Vinyals and Quoc Le left with him. Here is the plan.

Your Chatbot Now Has to Tell You It Is a Chatbot in the EU

EU AI Act Article 50 chatbot disclosure rules took effect on 2 August 2026. What counts as compliance, what does not, and the 15 million euro ceiling.

How Much Power Do AI Data Centres Actually Use, And Where Does It Come From

How much power do AI data centres use? A gigawatt in plain English, why sites are built next to power stations, and what the grid queue actually ...

OpenAI IPO valuation: what the confidential S-1 actually shows

OpenAI IPO valuation estimates run from $730bn to over $1tn. The confidential S-1 shows a negative 122% operating margin against a $25bn annual run ...

What is a Vision-Language-Action model, and why robots suddenly got useful

What is a vision language action model? One network takes camera input plus a spoken instruction and outputs motor commands, replacing the old ...

Big tech AI spending has hit 1.5 trillion dollars in commitments

Big tech AI spending in 2026 is nearing 1.5 trillion dollars in committed capacity, while around 80 percent of firms deploying AI report no profit ...

IBM just put OpenAI inside its consulting army

The IBM OpenAI partnership folds GPT-5.6 and Codex into IBM Consulting Advantage, giving OpenAI access to the enterprise last mile it has struggled to ...

Japan Has Launched the World's First National AI Infrastructure and the Scale Is Staggering

Most countries talk about building national AI capability. Japan has actually done it. Working with Noetra Corp.

The Open Secure AI Alliance's SAFE Framework Wants to Make Agentic AI Honest About Its Risks

As AI agents become capable of taking real actions in the world, sending emails, executing code, managing files, making purchases, the question of how ...

China Forces Meta to Unwind Its $2 Billion Manus AI Deal

Beijing ordered Meta to reverse its $2B acquisition of AI agent startup Manus, marking the first forced unwinding of a major AI deal.

The EU AI Act Is Now Being Enforced. Here Is What That Means

Europe's AI Office started enforcing the world's first comprehensive AI law on August 2, with fines up to 15 million euros for non-compliance.

Google Gemini Just Hit 1 Billion Monthly Users

Google's Gemini crossed 1 billion monthly active users, becoming its fastest-growing product ever while 63% of interactions now happen by voice.

Anthropic Will Watermark All Text Generated by Claude

Anthropic is embedding invisible watermarks into all Claude text output, making it the first major AI lab to watermark written content globally.

DARPA Just Flew a Real Combat F-16 Controlled Entirely by AI

DARPA's VENOM autonomy kit flew a standard F-16 fighter jet under full AI control at Eglin Air Force Base in June 2026.

Lovable Raises $400M at $13.3 Billion, Now Europe's Most Valuable Private AI Company

Stockholm-based vibe-coding startup Lovable closes $400M Series C, doubling its valuation to $13.3 billion in eight months.

OpenAI and Microsoft

21 new item(s): OpenAI and Microsoft; On the quantitative analysis of decoder-based generative models; A connection between generative adversarial ...

Toddler's tragic death from brain-destroying amoeba offers lessons for doctors

2 new item(s): Toddler's tragic death from brain-destroying amoeba offers lessons for doctors; The web’s newest weapon against AI scrapers is a font ...

Uncensor any LLM with abliteration

"DARPA's AI Just Flew a Real F-16, and a Human Could Take Over With One Switch"

"DARPA and the US Air Force flew a modified F-16 under AI control at Eglin Air Force Base, using the VENOM Autonomy Kit that lets a human pilot flip ...

EU AI Act Transparency Rules Are Live: Chatbots Must Now Identify Themselves

EU AI Act transparency rules went live on 2 August 2026, requiring chatbots, deepfakes, and AI-generated content to identify themselves to users.

DeepSeek Just Put Real Money Into Humanoid Robots. Unitree's $9 Billion IPO Shows Why.

Unitree Robotics priced its Shanghai IPO at $9 billion with DeepSeek investing $20.8 million for a stake in the humanoid robot maker.

Lumilens Raised 700 Million to Bring Optical Networking to the Data Center

Lumilens raised 700 million dollars to bring optical networking to the data center, swapping electrical wiring for light to break the AI bandwidth ...

OpenAI's Astra just solved 10 maths problems that stumped everyone for decades

An unreleased OpenAI model called Astra produced verified proofs for ten open maths and computer science problems, some unsolved for nearly 30 years ...

OpenAI Astra Solves Ten Open Maths Problems in a Single Day

OpenAI Astra cracked ten maths problems open for up to 30 years, with machine-checked Lean proofs, for about 2,000 dollars in tokens. Here is why it ...

Meta's Llama 4 Reaches 100 Million Users in Under Three Months

Meta's open-weight AI strategy has always been a bet that volume beats exclusivity.

Meta's Llama 4 Maverick Tops the Open-Source Leaderboard and Forces a Rethink

For most of the past two years, the story of open-source AI has been one of perpetual catch-up.

Why RAM prices are climbing fast in 2026, and AI is the reason

RAM prices have jumped sixfold in a year as memory makers chase AI demand. Here is why DDR5 costs more in 2026 and what it means for your next build.

An AI-Designed Vaccine Just Cleared Its First Human Trial

An AI-designed vaccine has cleared its first human trial. Here is what a vaccine whose active ingredient was created by AI passing early safety ...

An AI-Designed Vaccine Just Cleared Its First Human Trial

An AI-designed vaccine has cleared its first human trial. Here is what a vaccine whose active ingredient was created by AI passing early safety ...

Japan's National AI Infrastructure Launch Is a Blueprint Other Countries Will Copy

Japan has done something no other country had done before: launched a dedicated national AI infrastructure as a deliberate state-level project.

The EU AI Act's First Major Enforcement Action Has Arrived and It Targets Biometric Surveillance

The EU AI Act has had its first genuinely consequential enforcement moment.

What Pelé's Lost Goal Tells Us About AI and the Limits of Historical Truth

Let me take the DeepMind Pelé reconstruction one step further, because the project raises a question that deserves more than a paragraph.

Google DeepMind Used AI to Reconstruct a Pelé Goal Lost for 65 Years

There is a goal that Pelé scored in 1959 at the Rua Javari stadium in São Paulo that most people alive today have never seen. No broadcast exists.

Dark Matter Might Come in Two Flavours, Not One

A new theoretical model argues dark matter is at least two particles, not one, that collide with each other and could explain diffuse dwarf galaxies ...

"Dark Matter May Come in Two Particle Types, Not One"

"A new model says dark matter could be two kinds of particle that slowly drift apart inside galaxies, and it may explain several stubborn cosmic ...

Anthropic's 1.5 Billion Dollar Copyright Settlement Sets a New Precedent for AI and Creative Rights

Anthropic has just had a landmark 1.5 billion dollar copyright settlement approved by a US court, and it is the kind of moment that will be talked ...

Google DeepMind Rebuilt Pelé's Lost 1959 Goal Using AI

In 1959, a 19-year-old Pelé scored what witnesses described as one of the most extraordinary goals ever seen, a solo effort at Rua Javari in São ...

Netflix Spent 587 Million Dollars on Ben Affleck's AI Film Studio

Netflix has acquired Ben Affleck's AI filmmaking startup for 587 million dollars, and I have so many questions.

Dave Eggers Told OpenAI Staff That ChatGPT Is Silencing a Generation of Writers

There are not many people who would walk into OpenAI headquarters and tell the staff, to their faces, that their product is doing real harm to human.

EU's toughest AI Act rules take effect 2 August, fines up to 7% of turnover

Full enforcement of high-risk provisions under the EU AI Act begins 2 August 2026, with penalties that exceed even GDPR's maximum fines.

The EU Is Forcing Google to Open Up Search Data and Android AI

The European Union has made it official. Google must share its search data with rivals and open up AI features on Android to third-party developers.

Hyundai Factory Workers Strike Over Fear of Humanoid Robots

Workers at a Hyundai automotive factory have gone on strike, with their concerns centred specifically on the company's plans to deploy humanoid robots ...

xAI Can't Deny Grok Generates CSAM. So It's Suing Its Own Users

This is a story that should make everyone uncomfortable, and not just because of the subject matter.

Suno Secretly Scraped Millions of Songs From YouTube, Genius, and Deezer to Train Its AI

If you've ever uploaded a song to YouTube, posted lyrics to Genius, or streamed anything on Deezer, there's a decent chance your music ended up in ...

Apple's iOS 27 Public Beta Is Out and Siri Finally Feels Like It Means It

Apple released its public betas for iOS 27, watchOS 27, and macOS 27 this week, and if the early impressions from people actually running it are ...

PixVerse Raises 439 Million Dollars as AI Video Generation Heats Up

AI video generation startup PixVerse has raised 439 million dollars in a new funding round, pushing its valuation past two billion dollars.

Apple Sues OpenAI Over Alleged Trade Secret Theft

Apple has sued OpenAI over alleged trade secret theft, pointing to 400+ former staff now at OpenAI. Here is what the Apple OpenAI lawsuit really ...

Hugging Face's CEO Says the Era of Renting AI Is Ending

For the past few years, the dominant model for enterprise AI has been subscription access: you pay OpenAI, Anthropic, or Google a per-token fee to ...

OpenAI Is Going After Families and ChatGPT Is How

OpenAI has a new growth target, and it is not enterprises or developers. It is your household.

Apple Sues OpenAI Over Alleged Hardware Trade Secret Theft

The relationship between Apple and OpenAI has always been an unusual one.

Meta Kills Its AI Deepfake Feature After Public Backlash

It lasted about as long as you would expect. Meta has pulled a feature from Instagram that allowed users to generate AI images of public figures.

Fidji Simo Steps Down from OpenAI Due to Illness

Fidji Simo, who had been serving as one of OpenAI's most senior executives and was widely considered the company's number two, has stepped down from ...

Microsoft's Carbon Emissions Jumped 25 Percent Last Year. AI Is Why.

Microsoft has published its 2026 sustainability report, and the headline number is uncomfortable: the company's carbon emissions rose 25 percent in ...

OpenAI Launches GPT-5.6 and Names It the Preferred Model for Microsoft Copilot

OpenAI has released GPT-5.6, the latest entry in its model family, and has made a pointed declaration: this is now the preferred model powering ...

Netflix Pioneered Binge-Watching. Now Its Own Success Might Be Its Biggest Problem

There is a certain satisfaction in watching a company grapple with consequences of its own making.

AI Schools for Rich Kids: Inside the Tutoring Trend Splitting Education

There's a new status symbol among a certain set of American parents, and it's not a particular school or a tutor with Ivy League credentials.

Amazon Is Closing Mechanical Turk to New Users — and It's About AI

Amazon has announced it will stop accepting new customers for Mechanical Turk, its long-running crowdsourced labour platform.

Alibaba Has Banned Employees From Using Claude Code, and the Timing Tells a Story

Alibaba has reportedly told its employees they are no longer allowed to use Claude Code, Anthropic's AI coding assistant.

Claude Sonnet 5 Is Now the Default Model for Every Free and Pro User

Claude Sonnet 5 became Anthropic's default model for Free and Pro users on July 1, running agentic tasks near Opus 4.8 at a lower price.

Google Made a Fourth of July Ad About AI Helping Write the Declaration of Independence

Google released a Fourth of July commercial this week that depicts an AI assistant helping someone draft what appears to be an early version of the.

Google's AI Buildout Drove a 37 Percent Jump in Its Electricity Use Last Year

Google consumed 37 percent more electricity in 2025 than it did in 2024. That is not a number that appears buried in a footnote.

Midjourney Is Asking Hollywood to Come Clean About How It Uses AI

Midjourney has made an unusual move: the AI image generation company is asking Hollywood studios to publicly disclose how they are using AI in their.

Claude Sonnet 5 Is Now the Default Model for Every Free and Pro User

Claude Sonnet 5 became Anthropic's default model for Free and Pro users on July 1, running agentic tasks near Opus 4.8 at a lower price.

AI is quietly gutting entry-level tech jobs, and this week's jobs report shows the strain

Tech layoffs have hit 142,000 in 2026, with roughly 88,000 US cuts tied to AI. The roles vanishing first are the entry-level rungs, and that is a ...

Zuckerberg Admits Meta's AI Agents Are Behind Schedule

Mark Zuckerberg has told Meta staff that AI agents have not progressed as quickly as he had hoped, in a candid internal admission that cuts against ...

Digital Realty buys three data centres from Blackstone for $7.8 billion

Digital Realty is acquiring three fully-leased Northern Virginia data centres from Blackstone for $7.8 billion, a bet on sustained AI infrastructure ...

OpenAI is unveiling Codex hardware on 15 July

OpenAI will reveal new Codex hardware on 15 July 2026. Leaked images show a physical control panel form factor for an AI coding agent.

Claude Sonnet 5 is here, and it might be the most useful AI Anthropic has shipped

Anthropic's new mid-tier model launched on June 30. It's faster, more agentic, and cheaper than what came before.

Colorado's AI Act came into force on June 30. Here's what it actually requires

The first comprehensive US state AI law took effect this week. It targets automated decision-making in high-stakes contexts.

OpenAI just renamed everything. Here's what GPT-5.6 Sol, Terra, and Luna actually mean

OpenAI launched a new naming system with GPT-5.6 and three model tiers. We break down what Sol, Terra, and Luna are and who should use which.

80% of companies using AI report no measurable gains. The other 20% are pulling away fast

PwC's 2026 AI performance study found the top 20% of AI-exposed companies achieved 163% productivity growth. The bottom 80% got almost nothing.

The AI talent raid that won't slow down

Two more Gemini researchers joined Anthropic, the fourth senior Google exit in six days, as Anthropic accused Alibaba of millions of fraudulent Claude ...

DeepSeek just made its 75% price cut permanent

DeepSeek's V4-Pro now sits at $0.44 per million input tokens and $0.87 output, permanently. That is up to 17x cheaper than GPT-5.5.

GPT-5.6 slips to July as OpenAI goes quiet

GPT-5.6 was expected in late June. Prediction-market odds collapsed from 83% to 18%. The IPO quiet period looks like the reason.

OpenClaw went from 9,000 to 210,000 GitHub stars. Here is what it actually does

OpenClaw AI is the fastest growing open source project around. It runs a personal assistant on your own machine and wires it into WhatsApp, Slack and ...

OpenRouter: one key for every AI model

OpenRouter is a single gateway and API key that routes to hundreds of language models, with live pricing so you can compare cost per call.

OpenClaw went from 9,000 to 210,000 GitHub stars. Here is what it actually does

OpenClaw AI is the fastest growing open source project around. It runs a personal assistant on your own machine and wires it into WhatsApp, Slack and ...

Anthropic is using its own AI to help build the next AI

Anthropic is leaning into recursive self-improvement, using its models to help design their successors. We unpack the claim and why the trend matters.

ChatGPT falls below half the global AI assistant market

ChatGPT's share of the global AI assistant market dropped to 46.4 percent, its first time under 50 percent, as Gemini and Claude gained ground.

Coinbase for Agents lets your AI assistant spend real money

Coinbase launched a platform that lets AI assistants like ChatGPT and Claude trade, pay, and buy on your behalf. Here is what changes and what to ...

Databricks launches Unity AI Gateway to control runaway agent costs

Databricks introduced Unity AI Gateway, giving organisations spending limits, monitoring, and runaway-usage protection across AI providers.

The EU moved its biggest AI deadlines and banned nudifier apps

The EU Digital Omnibus package pushes back high-risk AI obligations, widens SME relief, and bans non-consensual image apps. What it means for ...

Zhipu AI releases GLM-5.2 under an open MIT licence

China's Zhipu AI shipped GLM-5.2, a large mixture-of-experts model with a one million token context window, under a permissive MIT licence.

Google's Gemini 3.5 Pro can now read 2 million tokens at once

Gemini 3.5 Pro is reaching general availability with a 2-million-token context window and a Deep Think reasoning mode, the largest context window in ...

ChatGPT's market share just fell below 50% for the first time

ChatGPT hit 46.4% of the AI assistant market in May 2026, the first time below 50% since its 2022 launch. Gemini is at 27.7%. Claude reached 10.3% ...

The man who helped invent every AI chatbot just joined OpenAI

Noam Shazeer co-invented the Transformer, built the architecture tricks that make AI cheap to run, and just left Google for OpenAI. Here is why it ...

The man who helped invent every AI chatbot just joined OpenAI

Noam Shazeer co-invented the Transformer, built the architecture tricks that make AI cheap to run, and just left Google for OpenAI. Here is why it ...

One in ten people now get their news from an AI chatbot

Weekly news use of AI chatbots jumped from 7% to 10% worldwide, says Reuters' 2026 report. The catch for publishers: most users never click the ...

GPT-5.5, Gemini 3.5 Flash and Claude Opus 4.8 all landed at once. Who is ahead?

Three new frontier models arrived close together in June 2026. Here is what the latest leaderboard reshuffle actually means for the people using them.

OpenAI filed to go public, and now two AI giants are racing to list

OpenAI confirmed a confidential IPO filing with the SEC, with bankers reportedly eyeing a September debut. Anthropic filed weeks earlier. Here is what ...

One in ten people now get their news from an AI chatbot

Weekly news use of AI chatbots jumped from 7% to 10% worldwide, says Reuters' 2026 report. The catch for publishers: most users never click the ...

MIT Startup Claims to Have Cracked a Core Mathematical Bottleneck Holding Back LLMs

Miami-based AI startup Subquadratic came out of stealth last month with a bold claim: it has solved a fundamental mathematical bottleneck that has ...

Built with AI this week: a machine-checked maths proof, a tutor that makes you type, and more

Our weekly roundup of genuinely impressive things people made with AI, all credited and linked. This week, an algorithm with a formal proof no human ...

Anthropic just gave everyone its most powerful AI, with the dangerous bits locked away

Claude Fable 5 is the first Mythos-class model the public can actually use. The catch? Hard safety limits, a steep price, and a data retention rule ...