Welcome back! Satya Nadella spent his Sunday warning that a handful of AI models could hollow out entire industries, which is a strange thing to hear from the man running one of them. Criminals stole almost £1.3 billion from people in Britain last year with AI doing the heavy lifting, so we lay out how to dodge the new scams. Mark Zuckerberg finally has the $14 billion model he paid Alexandr Wang to build, and now he has to sell it. And a review of 161 studies shows how AI is moving into the food on your plate.
In today's Generative AI Newsletter:
Microsoft: Why is the CEO of a top model maker warning about top model makers?
AI scams: Which AI-powered scams are draining accounts, and how do you stop them?
Meta: What does Zuckerberg have to prove now that the $14 billion model has shipped?
Food safety: How is AI learning to catch Salmonella before it reaches your kitchen?
Satya Nadella used a Sunday post on X to warn against a future where a few AI providers capture most of the value while everyone else loses ownership of their knowledge.
"There is no societal permission for an AI future that hollows out entire industries," he wrote, describing a world where "every company across every sector is ceding value to a few models that eat everything they see."
His comparison was globalization. Outsourcing hollowed out whole industrial economies, the GDP figures looked fine and the displacement was real. He wants companies to keep control of their own learning systems instead.
It is a pointed message from inside Microsoft, which builds some of those few models. The timing fits where the model makers are heading.
OpenAI is rebuilding ChatGPT into a superapp that runs other companies' software, and Anthropic bought a biotech and now sells a life-sciences Claude.
Nadella is not alone in the worry. Snowflake's CEO warned this year that software firms risk becoming a "dumb data pipe" feeding the model makers, and Box's CEO said a company's own context is the last real edge.
Why it matters: The companies that own the models are starting to do the work their customers do, and one of their own is naming where that road ends.
Want to get the most out of ChatGPT?
ChatGPT is a superpower if you know how to use it correctly.
Discover how HubSpot's guide to AI can elevate both your productivity and creativity to get more things done.
Learn to automate tasks, enhance decision-making, and foster innovation with the power of AI.
Google, Nvidia, Oracle and IBM sit among the partners following the GTM strategy dropping live tomorrow. The person who knows how companies at that level actually reach people is Steve Nouri, a Fortune 500 go-to-market advisor, and today he's handing over the playbook.
Distribution is what separates the AI companies everyone knows from the ones nobody finds. Whether you build, back or run one, this is how the top players do it.
This is for you if:
You want to put an AI product in front of millions, even if starting from zero.
You run a company and need to win more of your customers' attention without paying for it.
You back AI companies and want to spot the ones that can reach a market before you write the check.
Ask Steve your questions live and leave with the same playbook that reached 14 million people. It's the kind of access the big names usually keep in-house.
It's live tomorrow and the room is filling up. Save your seat here.
UK criminals stole almost £1.3 billion last year across 4.1 million cases where money vanished, up 11% on the year, and AI is the force multiplier.
UK Finance says fraudsters now clone voices, run longer romance cons and move faster than people can react. Here is how the main AI-driven scams work and how to shut each one down.
The cloned family voice: Three seconds of audio is enough to copy a voice, so a panicked call from your "son" or "mother" begging for urgent cash may be a machine. Agree a code word with close family and ask for it before you act on requests.
The romance build-up: Scammers run fake profiles for months on stolen photos, now backed by cloned voices on the phone, and one fraudster in the report married his victim to keep stealing.
The "safe account" call: Someone posing as your bank or the police tells you to move money to a "safe account" to protect it. No bank or police force will ask that, so hang up and call back on the number printed on your card.
The guaranteed return: Investment scam losses jumped 40% to a record, often dressed up with fake celebrity endorsements. Walk away from anything promising guaranteed or rushed returns and check the firm against your regulator's register first.
AI is built to rush you, so slow the moment down. Pause and verify on a channel you already trust before you send a penny.
A year ago Mark Zuckerberg paid more than $14 billion to bring in Alexandr Wang and his top Scale AI engineers and rebuild Meta's AI from the inside. The model exists, and now Zuckerberg has to turn it into money.
How Meta got here:
Llama 4 landed in April last year and fell flat with developers.
Two months later, Zuckerberg put $14.3 billion into roughly half of Scale AI and hired Wang.
Wang stood up Meta Superintelligence Labs to run the strategy.
This April, Muse Spark shipped, built to plug into Facebook, Instagram and the Ray-Ban glasses.
Wall Street is not sold. Meta stock is down 18% over the past year, worst in the megacap group, even after 33% revenue growth last quarter. Ads still bring 98% of the money, and developers stay cool on a model one startup CEO called a "yawn." Investors want AI-first products people pay for, and Muse Spark has not delivered that yet.
A new systematic review in npj Science of Food pulls together 161 studies on how fast AI is moving into food safety.
The work has exploded as labs use machine learning to spot pathogens, predict chemical contamination and trace outbreaks.
Microbiology leads at 35% of the research. In one case an "electronic nose" paired with classification algorithms identified Salmonella at 85 to 100% accuracy, and another predicted disease outcomes from raw Salmonella sequences at 87%.
One surveillance system used anonymized phone search and location data to find contaminated venues more than three times as effectively as standard investigations.
Most food samples are clean, so models rarely see the dangerous case, and privacy rules keep useful datasets locked away. The authors back explainable AI and federated learning to change food safety from cleaning up after outbreaks to heading them off.
Datagrid builds AI agents for the document grind of construction and engineering. Connect the systems a project already runs on and its agents read the files and do the work, reviewing submittals, answering RFIs and checking specs against drawings.
One customer cut spec review from three to four hours down to two to three minutes, and another reviewed eight submittals in an hour, work that took a four-person team a full day.
Try this yourself:
Sign up free at app.datagrid.com/sign-up, where the $0 plan lets you explore the agents.
Connect one system you already use, like Procore, Autodesk, SharePoint or Google Drive.
Hand an agent a real task, such as reviewing a submittal package or checking a spec section against the drawings.
Open the result and follow its citations back to the exact source file.
Who it's for: construction, engineering and other project teams buried in submittals, RFIs and spec reviews.
Washington pulled Claude Fable: As you’ve probably heard, Trump blocked access to Claude Fable and Mythos two days ago. Anthropic team is negotiating a “pardon” for their model to make sure everyone can access it again.
KT hires an AI translator: Korea's KT rolled out an "AI Multilingual Counselor" across its stores in more than 20 languages so foreign customers can sort out phone plans without the usual paperwork ordeal.
Geneva debates military AI: Diplomats took up the challenge of regulating AI in warfare at talks in Geneva, held alongside the G7, as the long effort to set limits on autonomous weapons grinds on.
AI cracked a crypto network: A researcher used Claude Opus 4.8 to surface a four-year-old Zcash flaw that could have minted unlimited counterfeit coins, and the token fell about 50% once the bug went public.
Canada wants AI built for everyone: The Canadian Human Rights Commission urged Ottawa to bake accessibility and equality into its new "AI for All" strategy and make sure diverse communities show up in the training data.
Learn more about AI from the experts building it
📸 Follow us on Instagram for fast, visual AI updates in 30 seconds.
🐝 Subscribe to our Atlas newsletter — trusted by 3M+ subscribers — to stay ahead of AI news across tech, education, and business.
📺 Watch us on YouTube to hear insights directly from leading AI voices, builders, and innovators.
🐦 Follow us on X for breaking AI news and real-time industry updates.
🧠 Learn how to build your next AI application with practical resources and expert guidance.
🚀 Explore investment opportunities in the future of AI and join our community-backed growth journey.
