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 2 of 76
Whoever Grants an AI Agent Access Is the One Governing It
Insurers have started asking for proof of AI controls, as they once did for ransomware backups, and that proof lives in what each agent is allowed to do, not in executive titles.
· 2 min read
Hackers Use Stolen AI Keys to Fund Their Own AI Attacks
Free AI hacking tools borrow paid AI models to run, so criminals now steal other people's AI keys, and any small business that builds on AI is in the sweep.
· 2 min read
California's OpenAI Suit Tests Who Answers for an AI Agent
California's lawsuit argues OpenAI is responsible for a hack its agents carried out on their own, the same logic a Canadian tribunal used in 2024 to make Air Canada pay for its chatbot's wrong advice.
· 2 min read
Alibaba's AI Got 67% Cheaper Per Task on a 25% Price Cut
Alibaba's Qwen4-Max cut the cost of a task by two thirds with a price-per-token cut of a quarter, because it writes far fewer tokens to finish the same job.
· 2 min read
AI Attackers Finish Before Anyone Notices, So Close Old Access
A security company's breach, finished in under two days through a vendor key it never switched off, shows that closing old access now protects better than watching for attackers.
· 3 min read
AI Sales Voices Win by Calling Back Fast, Not by Pitching Best
Jordan Belfort's AI sales voice still trails top closers, but leads go cold within hours, so a decent voice that calls back at once can beat a better salesperson who waits.
· 2 min read
Why 80% Accurate AI Works for Email but Fails at Month-End
An AI that is right 80% of the time is fine for drafting email, but chained across five finance steps it gets the whole job right only about one time in three.
· 2 min read
Workers Are Handing Eldercare Paperwork to AI, Out of Sight
One in six American workers looks after an aging relative and most say nothing at work, so the reading, calling and form-filling is moving to chatbots their employers never see.
· 3 min read
Frontline Headsets Used to Give Orders. Now Workers Can Ask
Warehouse voice systems have given workers orders since the early 1990s; AI lets workers ask open questions and get answers from company policy, so know-how moves from supervisors to the floor.
· 2 min read
AI That Paints Over Its Own Mistakes Can Finish Longer Jobs
Claude Fable 5.2 paints by writing code, looking at the canvas and covering its own errors, the same loop that lets AI finish long work without starting over.
· 2 min read
Australia's Early AI Job Signs Are Fewer Young Hires
The first signs of AI in Australian jobs are fewer young people being hired into software, clerical and call-centre work, much as cash machines cut the tellers each bank branch needed.
· 3 min read
Cheaper AI Does Routine Work and Calls a Smarter One When Stuck
Google's new Gemini 4 Hydrogen does the routine steps and hands the hard ones to a stronger model, and in Google's coding tool that pairing used 24% less quota than the strong model alone.
· 2 min read
When AI Makes the Deck, the Template Is the Real Work
Slide apps let an AI fill a company's own template from its sales records, so design and prices stay fixed and the person's job becomes owning the template.
· 2 min read
The Agent That Built the App Can Now Advertise It Too
Replit's agent now runs Google ads and buys supplies, and because it wrote the app it can add the tracking that shows whether an ad brought sign-ups, not just clicks.
· 3 min read
AI Is Creating New Ransomware Gangs, and Backups Aren't Enough
AI lets newcomers start ransomware gangs, and with many likely to steal files rather than lock them, a good backup no longer settles the threat.
· 2 min read
Surviving a Cyberattack Means Working With Your Systems Off
When a cyberattack forces a company to shut down its own systems, how much work it can do by hand decides the damage: Maersk kept 80% going, while M&S closed its online clothing shop for 46 days.
· 2 min read
When AI Writes the Work, the Person Who Sends It Owns It
A presenter blamed Claude for errors in a slide deck, but a tribunal already ruled that Air Canada owned its chatbot's wrong answer, so whoever sends AI-made work owns it.
· 2 min read
Claude Duet's Best Work Is Catching What Falls Between Apps
Anthropic's Duet agent is most useful when it spots that your email, calendar and Slack disagree, and least useful when it guesses how urgent something is.
· 3 min read
Meeting Notes Are Cheap Now, So Follow-Up Is What Counts
ElevenLabs' new work app is the latest to write up meetings and dictation, which makes notes the cheap part and leaves the value in whether the follow-up gets done.
· 2 min read
Hired AI Experts Lose Their Edge, So Employers Train Staff
AI tools change so fast that an outside hire has only months more experience than your own staff, so employers are choosing to train people who already know the business.
· 2 min read