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 11 of 76
Sharing a Meta Muse Chat Can Publish It for Anyone to Find
Muse builds each shared chat's web address from its title, and a public lookup list records that address, so the page can be found without the link.
· 2 min read
The Company That Watches AI Agents Should Not Sell Them
Microsoft's new Guardian Agents watch a company's other AI agents, and the Enron audit failure shows why the watcher should not also be the seller.
· 3 min read
AI Pays Off When Companies Plan What Staff Do Next
Few companies see big returns from AI, and the ones that do tend to decide what staff will do with the freed time, as Ikea did when it retrained 8,500 call-centre workers.
· 2 min read
Hiring One AI Expert Does Not Give a Board AI Oversight
A board can add an AI expert and still have nobody checking what its company's AI is allowed to do, a problem finance solved by making oversight something every director does and at least two do well.
· 2 min read
A Stolen AI Key Is a Credit Card With No Limit
Attackers who broke into an old OpenCode system ran up charges on users' OpenAI, Anthropic and Google accounts, because a stolen AI key spends freely until its owner sets a cap at the provider.
· 3 min read
The DeepSeek Ban Makes the Model Inside Software a Vendor Question
Banning DeepSeek from federal agencies also bans products built on it, so the AI model inside software becomes something every buyer and vendor must be able to name.
· 2 min read
AI Agents Are Making Website Break-Ins as Cheap as Spam
One hacker's AI agents broke into more than 100 company websites for roughly $50 each, which makes hacking work like spam: cheap enough to try on every site, not only the valuable ones.
· 3 min read
AI Pays Off When It Moves One Number Finance Already Tracks
AI that saves hours does not show up in a budget, so the projects that keep their funding are tied to a figure finance already watches, as computers were in the 1990s.
· 2 min read
A Forecast Is Not a Decision, and AI Makes That Matter More
AI hands anyone a forecast in minutes, so the valuable part is saying what a wrong guess costs in each direction, as with a baker who should bake above the average day.
· 2 min read
When AI Does the Doing, People Become the Bottleneck
Anthropic now hires far fewer junior engineers and its staff spend more time in meetings, because once AI does the work, the scarce thing is human attention and judgment.
· 3 min read
A Real Writer's Track Record Can Now Be Borrowed by a Fake
Someone using AI got a paid article into Science under a real journalist's name, which shows that checking a stranger's past work no longer proves who they are.
· 3 min read
Whoever Sends Out AI's Mistakes Is the One Who Pays
A tribunal made Air Canada pay for what its chatbot promised, and the same rule now reaches anyone who sends AI-written work to a client or customer under their own name.
· 3 min read
OpenAI's Agent Talked Itself Into Using Someone Else's Wallet
An OpenAI model decided on its own that using an account it found was fine, did the job, and left it out of its summary, so an agent's report is not a record of what it did.
· 2 min read
Workday Is Turning Its HR Software Into a Rulebook for Chat Apps
Workday is letting staff do HR and finance work from inside Claude, Copilot and Slack, which leaves the rules and approvals, not the screen, as the part worth owning.
· 3 min read
A Forgotten Password Is Now an Open Door for AI Agents
AI agents under test in India read confidential records using keys that businesses had left exposed, so old forgotten credentials are now a real weak point.
· 2 min read
Google's Deep Review Makes an Expert Second Read Cheap
Google's Deep Review, free for two documents a week, checks long contracts and reports for contradictions, so the careful second read that errors used to slip past is now cheap.
· 3 min read
ChatGPT Is Now Built on an Agent That Failed in the Cloud
OpenAI's coding agent flopped when it ran on a remote computer and took off once it ran on the user's own machine, which is why ChatGPT's desktop app is now built on it.
· 2 min read
Six-Cent Photo Edits Turn One Product Shot Into Many
Krea 3 edits a photo in seconds for $0.06, so a shop can photograph a product once and make its other versions from that picture, and the work left is checking them.
· 2 min read
Meta's New Terms Turn a Business Account Into AI Training Consent
From October 30, using Meta for business means agreeing to let it crawl your website and train AI on it, so a contract replaces a fair-use fight.
· 3 min read
Gmail Search Now Answers Questions Instead of Listing Emails
Gmail can now answer questions about your mail out loud, which does to search what Gmail's search once did to folders, and it makes the sender's wording the thing that gets read.
· 2 min read