Updated twice daily
Today’s top tech news
The ten stories shaping tech right now — researched, summarized, and explained so you know why each one matters.
Chinese humanoid-robot maker EngineAI files for a Hong Kong IPO
EngineAI, a three-year-old Shenzhen humanoid-robot startup last valued at about $1.5 billion, has filed confidentially for a Hong Kong IPO, working with CICC and Citic Securities. It opened a 12,000-square-metre factory on June 1 and began shipping its T800 robots — a line it says can turn out a humanoid every 15 minutes, geared for 10,000 units a year. EngineAI joins a stampede of Chinese robot makers, including AgiBot and PaXini, racing to list.
Why it matters: Humanoid robots are moving from demo videos to factory floors and public markets at once — and China’s makers are trying to win the category by manufacturing their way to scale.
SpaceX completes the largest IPO in history
SpaceX debuted on the Nasdaq under the ticker SPCX, pricing at $135 a share to raise roughly $75 billion at a valuation near $1.77 trillion — the biggest public offering ever, with the stock closing up about 25% in early trading. The S-1 revealed how deep the company has gone into AI: a February all-stock acquisition of Elon Musk’s xAI, $18.67 billion in 2025 revenue, and roughly $30 billion in GPU-compute deals with Google and Anthropic via its Colossus data centers.
Why it matters: It’s the clearest sign yet that ‘space company’ and ‘AI infrastructure company’ are merging — and that the public markets are now where the AI build-out gets funded.
NVIDIA pitches its Vera CPU to China to skirt US export controls
With its AI GPU sales to China still frozen, NVIDIA opened orders for its 88-core Arm-based Vera CPU to Chinese cloud customers, with deliveries as early as August — and Alibaba and ByteDance already on board. The Vera is the CPU half of NVIDIA’s Vera Rubin platform; a single chip runs north of $20,000, and a full 256-chip rack about $10 million. The move exploits a gap in US rules, which haven’t restricted advanced CPUs the way they have AI GPUs.
Why it matters: It’s NVIDIA finding a side door back into a market worth tens of billions — and a live test of how long export controls hold when the rules cover GPUs but not the CPUs feeding them.
OpenAI acquires Ona to run Codex agents for hours in the cloud
OpenAI agreed to acquire Ona — the German startup formerly known as Gitpod — which runs AI agents in cloud-based sandboxes. The deal folds Ona’s team and technology into Codex, letting the coding agent take on tasks that run for hours or days without being tied to a single machine or session. OpenAI says more than 5 million people now use Codex each week, up 400% from earlier this year; terms weren’t disclosed and the deal is subject to regulatory approval.
Why it matters: Coding agents are only as useful as the environment they run in. Buying the cloud sandbox is OpenAI’s move to make Codex a long-running coworker rather than a one-shot autocomplete.
Quantum computing strings together a breakout month
June brought a run of concrete quantum milestones rather than the usual promises. A Zhejiang University team demonstrated a working quantum random-access memory (QRAM) architecture on a superconducting processor, Atom Computing showed the first continuous multi-round error correction on a neutral-atom system, and ORCA Computing moved photonic quantum hardware into a commercial London data center with Digital Realty. The throughline is access: cloud programs and lab integrations are putting real quantum machines in front of ordinary researchers.
Why it matters: Quantum has spent years as a someday technology. A month of error-correction and data-center milestones is what the transition from physics demo to usable infrastructure actually looks like.
The 2026 World Cup is normalizing AI mass surveillance
As the World Cup opens, US stadiums including Gillette, Hard Rock, and Mercedes-Benz are rolling out AI facial recognition for entry and payments, while camera-equipped robot dogs patrol venues in Dallas and New Jersey and counter-drone systems guard the airspace. Boston Dynamics says its robots don’t do facial recognition, but privacy groups are alarmed: the ACLU, Privacy International, and the EFF warn that defense contractors are using the tournament as a global showcase to normalize battlefield-grade surveillance — infrastructure that tends to outlast the event that justified it.
Why it matters: Big events are where surveillance tech gets its civilian debut. What’s deployed to watch a soccer match rarely gets uninstalled when the match ends.
Jeff Bezos’s Prometheus raises $12B at a $41B valuation for physical-world AI
Jeff Bezos’s stealthy AI startup Prometheus — which he co-leads with Google veteran Vik Bajaj — raised $12 billion at a $41 billion valuation from investors including JPMorgan, BlackRock, Goldman Sachs, DST Global, and Arch Venture Partners. The company is building AI tools to speed up the engineering and manufacturing of physical products — robots, jet engines, drugs, chip design, and data centers — by 10x or more. It is one of the largest rounds ever for a months-old startup.
Why it matters: Most AI money chases software; Prometheus is a bet that the bigger prize is AI that designs and builds physical things — and that a name like Bezos can raise the compute to try.
Visa plugs its payment network into ChatGPT, letting AI agents shop and pay
Visa and OpenAI announced a partnership at Visa’s Payments Forum that embeds Visa’s payment rails directly inside ChatGPT, so AI agents can complete real purchases on a user’s behalf. Transactions run within user-set guardrails — spending limits, merchant restrictions, and approval rules — with tokenized credentials and real-time fraud monitoring. The deal also covers developer and enterprise workflows powered by Codex, and follows OpenAI quietly retiring its earlier Instant Checkout feature in March.
Why it matters: Agentic commerce only works if an agent can actually pay. Wiring the world’s largest card network into ChatGPT is the piece that turns ‘an agent that shops’ into ‘an agent that buys.’
OpenAI weighs steep price cuts to fend off Anthropic
OpenAI is considering significant cuts to its token prices to pull customers back from Anthropic, the Wall Street Journal reported. The pressure is real: Anthropic’s revenue has surged on the back of Claude Code, and its $65 billion Series H pushed its valuation to $965 billion — above OpenAI’s $852 billion for the first time. Both companies are weighing the move just as they prepare for public listings, with OpenAI said to burn cash at many times its rival’s rate.
Why it matters: A price war between the two frontier labs would ripple straight to every developer’s bill — and signals that, at the top, the AI race is now as much about unit economics as raw capability.
Meta lays off about 8,000 — 10% of staff — in an AI-funding restructuring
Meta began notifying roughly 8,000 employees — about 10% of its workforce — that they are being laid off, the first wave of a restructuring it frames as necessary to fund its AI push. Singapore-based staff were the first to learn, with emails arriving at 4 a.m. local time, followed by the UK and US. The company also reassigned about 7,000 employees to AI initiatives, and Mark Zuckerberg told staff no further broad cuts are expected this year. Meta’s 2026 capital spending runs to as much as $145 billion.
Why it matters: It is the starkest version yet of the trade every big tech company is making: cut headcount to fund compute. The people are the budget line that pays for the GPUs.
Google’s Nano Banana Pro can finally render readable text in AI images
Google made Nano Banana Pro (Gemini 3 Pro Image) generally available — an image model built with the reasoning of Gemini 3, and notably the first where a prompt like ‘add the text Sale in bold white on the product’ reliably produces legible words instead of decorative gibberish. A lighter sibling, Nano Banana 2, even accepts a video file or YouTube URL as context to generate thumbnails and posters. The release lands as Google retires its older Imagen models in favor of the Gemini line.
Why it matters: Readable text has been the embarrassing failure of AI image generation for years. Fixing it turns these models from toy-art generators into tools that can actually make a usable poster, ad, or thumbnail.
HHS will use AI built on ChatGPT to audit all 50 states for fraud
The U.S. Department of Health and Human Services launched AERO — Audit Enforcement and Risk Oversight — a program that uses AI tools, reportedly built in part with ChatGPT, to scan at least five years of audit records across all 50 states for fraud, waste, and unresolved findings. Officials estimate $100 billion to $200 billion in wasteful or fraudulent spending each year and have put every state governor and treasurer “on notice,” with loss of funding as the penalty for chronic noncompliance.
Why it matters: It’s one of the most consequential government deployments of generative AI yet — a federal agency pointing ChatGPT-class tools at how every state spends billions in public money.
OpenAI confidentially files for a U.S. IPO that could top $1 trillion
OpenAI has submitted a confidential S-1 to the SEC, confirming the move on June 8, with Goldman Sachs and Morgan Stanley leading and Citigroup and JPMorgan in talks to join. Valued above $850 billion and generating roughly $2 billion a month, the ChatGPT maker is eyeing a public debut as soon as September at a valuation of up to $1 trillion. The filing lands a week after Anthropic’s own confidential submission and days before SpaceX begins trading.
Why it matters: A near-trillion-dollar listing would be one of the largest IPOs ever — and the clearest sign yet that the AI build-out is graduating from private megarounds to the public markets.
Anthropic publicly releases Claude Fable 5, its most capable model yet
Anthropic launched Claude Fable 5, the first publicly available version of its Mythos-class model, just days after warning that AI is becoming dangerously capable. The company says it sets state-of-the-art marks across software engineering, knowledge work, and vision, priced at $10 per million input and $50 per million output tokens — less than half its preview. In high-risk areas like cyber, bio, and chemistry the model refuses and falls back to Claude Opus 4.8, with the safeguards triggering in under 5% of sessions.
Why it matters: It is the first time Anthropic has shipped frontier Mythos-class capability to everyone — a live test of whether you can release maximum capability and hard safety limits in the same product.
Apple’s rebuilt Siri will run on Google’s Gemini
At WWDC 2026 — Tim Cook’s farewell keynote — Apple unveiled a completely rebuilt assistant, “Siri AI,” that leans on a custom, Apple-tuned version of Google’s Gemini running in Apple’s data centers under a reported $1-billion-a-year deal. On-device tasks like dictation and on-screen awareness stay on Apple Silicon; heavier reasoning routes to the Gemini-powered cloud via Private Cloud Compute. Siri also gets its own dedicated app, with a beta arriving later this year alongside iOS 27.
Why it matters: Apple paying its biggest rival to power Siri is a stunning admission of where the AI race stands — and the clearest sign that even Apple couldn’t go it alone on frontier models.
SpaceX reveals “AI1,” an orbital data-center satellite, ahead of its IPO
Elon Musk unveiled AI1, the first generation of SpaceX’s orbital AI compute platform — a solar-powered satellite with a 70-meter wingspan that sustains about 120 kilowatts of compute (peaking near 150) and sheds heat through a liquid radiator into space. Musk called it “much simpler than a Starlink satellite,” with two prototypes targeted for early 2027 and a constellation to follow. The reveal lands the same week SpaceX prices its IPO, with shares set to debut on the Nasdaq under the ticker SPCX.
Why it matters: If compute can run cheaply in orbit on free solar power, the data-center crunch on Earth gets an escape hatch — and SpaceX turns its launch dominance into an AI-infrastructure one.
NVIDIA launches Rubin — six new chips and a next-generation AI supercomputer
NVIDIA introduced Rubin, the platform succeeding Blackwell, unveiling six new chips and a rack-scale Vera Rubin NVL72 system built for training and inference at frontier scale. The company said the major clouds — AWS, Google Cloud, Microsoft Azure, and Oracle — plus specialist providers like CoreWeave and Lambda will be among the first to deploy Rubin-based instances, with Microsoft slating it for its next-generation “Fairwater” AI data centers.
Why it matters: Rubin resets the ceiling on AI compute again and locks every major cloud into NVIDIA’s roadmap for another generation — the chip cycle, not the model, still sets the pace of the whole industry.
Google launches tools to govern AI agents in the enterprise
Google introduced new controls for running AI agents at scale, including Agent Identity, which extends access management to agents the way it works for human users, and Agent Gateway, which enforces policy on every agent-to-agent and agent-to-tool connection. The push treats autonomous agents as first-class actors that need identity, permissions, and an audit trail — not as unmanaged scripts wired into production.
Why it matters: As companies deploy fleets of AI agents, the hard problem shifts from building them to governing them — and whoever owns agent identity and access becomes the control plane for enterprise AI.
Google makes Gemini 3.5 Flash the default for AI Mode in Search worldwide
Google has rolled out Gemini 3.5 Flash as the new default model behind AI Mode in Search, for everyone, globally. The company says it delivers sustained frontier-level performance for agentic and coding tasks at a fraction of the latency of its larger siblings. Google is pairing the change with new information agents arriving first for AI Pro and Ultra subscribers this summer.
Why it matters: Putting a frontier-grade model in front of Search’s billions of users by default is how AI answers stop being a separate tab and become the way search works.
Hackers slipped password-stealing malware into Microsoft’s open-source dev tools
Microsoft cut off access to dozens of its open-source GitHub projects after attackers injected credential-stealing malware into the code. Many of the affected tools relate to Azure and to AI coding workflows — used alongside apps like Claude Code, Gemini’s CLI, and VS Code — and the malware harvested developers’ passwords and tokens when the compromised tools ran. Researchers who flagged it suspect a link to an earlier breach of Microsoft’s Durable Task project in May.
Why it matters: It is a textbook software-supply-chain attack aimed squarely at AI developers — a reminder that the tools we build with are now a primary target.
The AI data-center boom hits a wall of transformers and permits
Roughly half of the AI data-center capacity planned for 2026 in the U.S. is now expected to slip or be cancelled, with only about 5 of 12 announced gigawatts under active construction. The bottleneck isn’t chips but the grid: substation-transformer lead times have stretched past 160 weeks, and permitting and interconnection add years after approval. The squeeze persists even as Alphabet, Amazon, Meta, and Microsoft pour a combined $650 billion-plus into capacity this year.
Why it matters: Capital is no longer the constraint — physics and the power grid are. The next AI cycle may be won by whoever can actually get electrons to the GPUs.
Supabase raises $500M at a $10.5B valuation as “vibe coding” takes off
The open-source database startup nearly doubled its valuation in eight months, closing a $500M Series F led by GIC with Stripe, Accel, Y Combinator and Salesforce Ventures. Supabase says launches on its platform are up more than 600% in a year — and over 60% are now spun up by AI coding tools, not people.
Why it matters: It’s the clearest sign yet that AI-written apps are reshaping who — and what — builds the back end of the internet.
Google pledges to make its AI data centers “water-positive” by 2030
Facing local backlash over AI’s thirst, Google became the first hyperscaler to vow it will replenish more water than its data centers consume by 2030. The plan leans on air cooling in stressed watersheds, recycled water, public reporting and $500M for water infrastructure, with 165 projects already running.
Why it matters: Water, not just electricity, is becoming the resource — and the public-relations battleground — that limits the AI build-out.
Generalist AI raises $400M at a $2B valuation for general-purpose robots
Backed by NVIDIA and Bezos Expeditions, the startup from ex-DeepMind scientist Pete Florence raised $400M led by Radical Ventures to build foundation models for robots. Its GEN-1 model, trained on 500,000+ hours of real-world data, claims dexterous tasks at 99% reliability and up to 3× past speeds.
Why it matters: Money is flooding into “physical AI” on a bet that the next foundation-model race is won in the real world, not the chat box.
AI music maker Suno raises $400M at a $5.4B valuation — lawsuits and all
Suno more than doubled its valuation in seven months with a $400M Series D led by Bond Capital, even as major labels press copyright suits over tens of thousands of songs. The platform now counts 2M+ paying subscribers and says users generate over 7 million tracks a day.
Why it matters: Investors are betting generative music wins in the market faster than the courts can rule on whether it is legal.
GitLab cuts 14% of staff and exits 22 countries to bet on the “agentic era”
GitLab is laying off about 350 people and pulling out of 22 countries, flattening management to refocus on AI that writes, reviews and manages code. Even after the cuts — costing $30–35M — it posted Q1 revenue of $264M, up 23%, and plans to plow most savings into its AI products.
Why it matters: A profitable, growing developer-tools company restructuring around agents shows how deeply AI is reshaping software work itself.
NVIDIA and SK hynix sign a multiyear pact for next-gen AI memory
NVIDIA and SK hynix announced a multiyear technology partnership to co-develop the memory behind the AI buildout, targeting HBM4 for NVIDIA’s Vera Rubin platform with initial deliveries planned for Q3 2026. SK hynix will also build memory for Vera CPUs, RTX Spark PCs, and Jetson Thor robotics computers, extending a co-engineering relationship that already powers NVIDIA’s top AI systems.
Why it matters: Memory bandwidth, not raw compute, is increasingly the bottleneck for AI. Locking in SK hynix for HBM4 is NVIDIA securing the scarcest ingredient of its next chip generation.
Microsoft unveils its own “MAI” models to cut its OpenAI dependence
At its Build conference, Microsoft launched a family of seven in-house models led by MAI-Thinking-1, a from-scratch reasoning model, and MAI-Code-1-Flash, which turns plain-English descriptions into working code. AI chief Mustafa Suleiman said the models beat OpenAI’s GPT-5.5 on internal benchmarks while costing up to 10× less to run.
Why it matters: Microsoft building credible models of its own loosens OpenAI’s grip on the world’s biggest enterprise software channel.
ChatGPT crosses 1 billion monthly users — the fastest app ever to get there
ChatGPT’s app passed one billion global monthly active users about three years after launch, reaching the milestone faster than Google Maps, TikTok, Instagram or YouTube, per Sensor Tower estimates. Google’s Gemini trails at roughly 900 million, with Claude and Grok far smaller at about 56 and 50 million.
Why it matters: A billion-user habit cements ChatGPT as the default front door to AI — and a powerful distribution moat for OpenAI.
NVIDIA’s Isaac GR00T puts a ready-to-train humanoid in researchers’ hands
At GTC Taipei, Jensen Huang unveiled a reference humanoid that pairs Unitree’s H2 Plus body and Sharpa five-fingered hands with NVIDIA’s Jetson Thor compute and open Isaac GR00T models — a single robot labs can buy, plug in and train. Ai2, ETH Zurich, Stanford and UC San Diego are first in line, with units shipping from Unitree late this year.
Why it matters: Turning humanoid robotics into an off-the-shelf platform could do for robots what reference designs once did for PCs.
SpaceX targets a record $75B IPO at a $1.75 trillion valuation
SpaceX kicked off its roadshow for a June 12 Nasdaq debut under ticker SPCX, pricing 556.6 million shares at $135 to raise about $75 billion — the largest IPO in history. The deal would value the company near $1.75 trillion, above Tesla, even as Morningstar pegs fair value at less than half that.
Why it matters: A successful listing would hand Musk an unrivaled war chest for Starship and Starlink — and set a new ceiling for what a private company can be worth.
Meta is racing to build AI “data centers” that are basically giant tents
To add compute faster, Meta is erecting fabric-wrapped “rapid deployment structures” — borrowed from Tesla’s playbook — at sites in Ohio and Tennessee, each roughly 125,000 sq ft and standing in about three months. Its Prometheus campus runs on 400MW of on-site gas turbines, sidestepping the grid entirely, as part of a 2026 capex bet topping $125 billion.
Why it matters: When the bottleneck is power and time, even Meta will trade polished data halls for tents to keep its AI buildout moving.
TSMC warns the AI chip shortage will last years — and 3nm prices are going up
CEO C.C. Wei said global advanced-chip supply will stay short of AI demand, with 2nm capacity already sold out through 2027. TSMC is weighing 3nm price increases of up to 15% in the second half of 2026, with another 5–10% possible in 2027.
Why it matters: Every Nvidia, AMD and hyperscaler chip starts at TSMC, so its pricing and scarcity ripple through the entire AI economy.
Cisco scrambles on an unpatched, actively exploited SD-WAN zero-day
Cisco disclosed CVE-2026-20245, a flaw in Catalyst SD-WAN Manager that lets a low-privileged attacker run commands as root by uploading a crafted file — its seventh exploited SD-WAN zero-day of 2026. Mandiant reported the bug; no patch is available yet, and Cisco has seen attackers push config changes to edge devices.
Why it matters: SD-WAN controllers sit at the center of corporate networks, so a root-level zero-day with no fix is a worst-case foothold for intruders.
Trump signs an executive order on “Advanced AI Innovation and Security”
The June 2 order pairs hardening government and industry systems against AI-enabled threats with a voluntary framework under which developers can give the government up to 30 days of pre-release access to frontier models. Agencies face aggressive deadlines, with the first deliverables due July 2.
Why it matters: It signals Washington’s preference for voluntary, security-focused AI oversight over hard regulation — shaping how frontier models get vetted.
Anthropic extends its vulnerability-hunting Mythos AI to the EU and NATO
Anthropic expanded Project Glasswing — its controlled-access program around Claude Mythos, a model it says has autonomously found over 10,000 critical zero-day vulnerabilities — to about 150 organizations across 15+ countries, including Samsung, SK Hynix and NATO. The EU cyber agency ENISA became the first EU institution admitted, ending a tense transatlantic standoff over access.
Why it matters: An AI that can mass-discover zero-days is dual-use dynamite, so who gets it — and under what safeguards — is now a geopolitical question.
Blue Origin’s New Glenn explosion leaves NASA’s moon plan leaning on SpaceX
After a New Glenn rocket disintegrated during a pad test at Cape Canaveral, NASA is urging Blue Origin to find a new launcher for its Blue Moon lander to keep Artemis on schedule. Blue Origin vows to resume flights by year’s end, but with its lander sidelined, NASA’s return to the Moon now hinges largely on SpaceX’s Starship.
Why it matters: Losing a second heavy-lift provider concentrates America’s lunar ambitions — and leverage — in Elon Musk’s hands just as SpaceX goes public.
OpenAI’s “Dreaming V3” gives ChatGPT a far better memory
OpenAI rolled out its biggest memory overhaul since ChatGPT first launched: a new architecture that quietly synthesizes your preferences and context in the background, reportedly using about 5× less compute. It’s reaching Plus and Pro users in the US first, with lighter memory landing on the free tier too.
Why it matters: Persistent, low-cost memory is what turns a chatbot into an assistant that actually knows you — and reopens hard questions about what it should remember.
Apple’s WWDC 2026 lands June 8 — and a chatbot Siri is the headline
Ahead of Monday’s keynote, Apple is expected to unveil iOS 27 with a fully conversational, ChatGPT-style Siri that can see what’s on your screen and act across apps. No new iPhone hardware is expected, but the software groundwork for a rumored foldable may be on display.
Why it matters: After a rocky “Apple Intelligence” rollout, a genuinely capable Siri is Apple’s chance to prove it isn’t falling behind in AI.
NVIDIA’s RTX Spark wants to put an Arm superchip in your next laptop
At Computex, NVIDIA unveiled RTX Spark — an Arm-based laptop chip built with Microsoft that pairs multi-core CPUs with Blackwell-class graphics and on-device AI. Laptops are expected around autumn 2026, with Adobe rebuilding Photoshop and Premiere Pro to run natively.
Why it matters: It’s NVIDIA’s most direct shot yet at Intel, AMD and Qualcomm in the Windows laptop — and a bet that every PC becomes an AI PC.
Anthropic files confidentially for an IPO at a ~$965B valuation
Anthropic submitted a confidential S-1 to the SEC, reportedly showing a $47B revenue run-rate and a $965B post-money valuation from its Series H round. Analysts increasingly treat a trillion-dollar debut as the base case.
Why it matters: It would be one of the largest tech listings ever — and a milestone test of how public markets value frontier-AI labs.
SpaceX signs a $920M-a-month compute deal with Google
SpaceX agreed to lease roughly 110,000 Nvidia GPUs from Google for about $920 million per month through mid-2029. The deal underscores how even cash-rich companies are renting, not buying, their way to AI scale.
Why it matters: Compute is the new oil — and multi-billion-dollar leases like this show just how scarce and strategic GPUs have become.
Congress drops a 269-page AI bill that could override state laws
The proposed “Great American AI Act” would preempt state AI rules for three years and impose safety-incident reporting and disclosure requirements on companies with $500M+ in revenue, plus a new federal AI standards center. It sets up a direct clash with laws like Colorado’s, which takes effect June 30.
Why it matters: A single federal framework could reshape AI rules nationwide — and decide whether states keep any power to regulate.
Alibaba’s Qwen3.7-Plus brings a 1M-token context at $2 per million
Alibaba released Qwen3.7-Plus, a multimodal model with a one-million-token context window priced at roughly $2 per million tokens — about 60% cheaper than its predecessor. It’s another aggressive move in the price war among frontier models.
Why it matters: Cheaper, longer-context models keep collapsing the cost of building AI products — great for developers, brutal for margins.
Qualcomm unveils a full-stack robotics platform for humanoids
Qualcomm announced its Dragonwing IQ10 series, a full-stack architecture meant to power everything from small household robots to full-sized humanoids. It lands as embodied-AI startups raise huge rounds — Generalist just closed $400M at a $2B valuation.
Why it matters: The race to put AI into physical robots is heating up, and the chips underneath it are becoming a battleground of their own.
Anthropic floats a coordinated “pause” if AI starts improving too fast
Anthropic urged leading labs to agree on shared protocols to slow development if AI systems begin self-improving faster than society can adapt. The proposal lands as rivals push ever more capable agentic models.
Why it matters: It reframes safety as an industry-wide coordination problem — and tests whether fierce competitors would ever agree to ease off the gas.
UK forces Google to let publishers opt out of AI Overviews
The UK’s competition regulator will require Google to let publishers exclude their content from AI-generated search summaries without losing normal search visibility. It’s one of the first concrete remedies aimed at AI’s impact on the open web.
Why it matters: AI summaries are siphoning traffic from the very sites they’re built on — and this is an early model for how regulators might push back.