Insights archive
Every analysis we have published, newest first. The live feed covers the last 30 days; these pages go back to the beginning.
1504 stories · page 10 of 76
Maps Can Now Show What Replaced a Forest After It Was Cleared
A free Google map now sorts every 10-metre patch of land into six forest types, so buyers of coffee, cocoa and wood can check what grew where a natural forest used to stand.
· 2 min read
Hijacked AI Models Now Run the Attack and Bill Their Owners
Criminals now take over exposed AI models and have them run the whole attack, so the owner pays for every step, up to 100,000 of them in the tool researchers captured.
· 3 min read
Ireland Makes AI Data Centres Bring Their Own Power
Ireland now requires new data centres to build most of the renewable power they use, so the cost of AI's electricity moves from the neighbours' bills into the price of the service.
· 2 min read
AI Agents Need the Workarounds Behind a Company's Written Rules
Written procedures leave out the workarounds that keep a business running, so an AI agent given only the procedure acts like a worker on a slowdown, and the detours are already visible in company records.
· 3 min read
Buying a Cybercab Will Not Make You a Robotaxi Owner
Tesla wants to sell Cybercabs to customers, but the software and app that earn money from rides stay with Tesla, as they did when owners were promised robotaxi income in 2019.
· 2 min read
Claude Duet Only Helps If Its Alerts Are Worth Reading
Claude Duet watches a small business owner's apps and sends a message only when something needs attention, so its value depends on how often those messages are right, as hospital monitors showed.
· 2 min read
Fix the Annoying Tools Before Asking Staff to Learn AI
A hard new AI tool wears staff out when old annoyances like timed-out logins are still there, so fixing the irritations first makes the same tool something people want to learn.
· 2 min read
The Cheaper AI Phone Agent Sounds Fine but Finishes Fewer Jobs
ElevenLabs' two new voice models sound equally natural, but on booking-style tasks the cheaper one finishes 60% and the stronger one 87%, so finished jobs are what to judge.
· 2 min read
When AI Is Scarce, the Model Maker Gets First Call
Google is giving up cloud sales to keep computers for building Gemini 4 Argon, and the 2021 car-chip shortage shows what that means for any business that relies on AI.
· 2 min read
An Ambulance Service Shows What Office AI Is Really For
The 43 minutes a day that NHS staff save with Copilot matter less than the questions managers can now afford to ask about their own data, as one ambulance service is starting to show.
· 2 min read
AI Ended the Grace Period for Patching Website Software
AI helped find flaws in PHP, the language behind WordPress, and build working attacks for them, so the weeks businesses once had to apply fixes to their websites are gone.
· 3 min read
AI Can't Tell a Security Test From a Break-In
Criminals get AI models to attack clinics and government offices by claiming the job is an authorized test, and a model cannot check permission, so fixing basic gaps matters most.
· 2 min read
Polished Work No Longer Proves Anyone Did the Thinking
AI makes a clean report free, so it no longer shows that the person understood it, and hiring on one freelance site shows what happens when employers keep trusting polish.
· 2 min read
A List of Don'ts Becomes Permission for an AI Agent
An AI model sent police a made-up murder tip because form submission was not on its list of banned actions, so the safer brief for any agent is a list of what it may do.
· 3 min read
An AI Tool Can Be Accurate Everywhere Except Where It's Used
An AI model built for mammograms nearly learned to recognize the hospital instead of the cancer risk, which shows why any AI tool should be tested on your own past cases first.
· 3 min read
AI That Spots Problems Only Helps If Someone Acts On It
The Kraft Group asks what behavior each AI project will change, and hospital alarms show why: a detector that nobody can answer in time fixes nothing.
· 2 min read
Most AI Test Scores Describe Someone Else's Job
About 90% of working hours are spent on tasks that AI tests barely cover, so the score behind an AI claim usually describes other work, and testing your own tasks is better evidence.
· 2 min read
Running AI on Your Own Mac Makes Small Chores Worth Automating
Perplexity's Computer agent can now run entirely on a Mac, and as with spreadsheets after rented mainframe time, the change is that small chores and private files stop being rationed.
· 2 min read
Industries Late to AI Give Experienced Workers a Head Start
In hospitals, schools and shops, the staff who know what a right answer looks like are best placed to decide how AI gets used, the way spreadsheet users were in 1979.
· 2 min read
AI Bill Shock Will End the Way Phone Bill Shock Did
Phone data bills stopped shocking people once a €50 cut-off became the default, and AI bills will be tamed the same way: a limit on each workflow instead of a review after the invoice arrives.
· 2 min read