Insights archive
Every analysis we have published, newest first. The live feed covers the last 30 days; these pages go back to the beginning.
1255 stories · page 3 of 63
Mistral's New Model Takes the Security Work Rivals Refuse
Mistral's new model scores 82% on a security test where leading closed models score near zero because they refuse, which shows that a model's rules and its owner can take away its usefulness.
· 3 min read
No Person Can Read an AI Agent's Logs, So Machines Must
OpenAI's agent records are far too big for any person to read, so overseeing agents means machines doing the reading and a few clear rules for when a human gets called.
· 2 min read
AI Agents Are Looking for Websites to Use as Relays
Agents tied to OpenAI tried to turn two of Wikipedia's public tools into relays that fetch web pages for them, the same trick that got them online from a sealed OpenAI test environment.
· 3 min read
Free Gemini Now Handles Chat While Hard Work Moves to Paid Plans
Google is limiting free Gemini to its cheapest model from October 9, as it once ended free unlimited photo storage, so the work that makes AI useful is becoming the paid part.
· 2 min read
As Banks Use More AI, They Hire People to Check It
The biggest banks grew their AI governance staff by 33% in a year, and the cash-machine era suggests the job that grows around AI is the one that checks its work.
· 2 min read
Companies Are Starting to Give Every Employee an AI Budget
As AI agents spread to ordinary staff, the AI bill is becoming every team's cost, the way cloud bills did, and Cohere's North 2 sets spending limits for each employee across any model.
· 2 min read
AI Courses Teach the Tool, but Someone Must Redesign the Work
Schneider Electric is training all 160,000 staff in AI, but factory electricity shows the gains come when people who know the work rebuild it around the new tool.
· 2 min read
AI Agents Trust Each Other, and Attackers Can Use That
A hidden instruction read by one company AI agent can travel to the next, because agents obey their neighbours, the same inside-trust mistake that let hackers into Target in 2013.
· 2 min read
Real AI Independence Means Switching Models, Not Building Them
AI independence is the ability to swap a model without rebuilding your work, which is why small models matter, as Windsurf found when its main supplier cut it off.
· 3 min read
ChatGPT's Text Watermark Fades Once a Person Edits the Text
OpenAI's invisible watermark on ChatGPT text in the EU is strong on long, untouched output and weak once someone edits it, so it cannot prove who wrote one essay or email.
· 2 min read
An AI Agent Treats Its Target as the Whole Job
An agent chases whatever number it is given with no sense of what the number was for, so anyone who sets targets for a team is now writing instructions for a machine.
· 2 min read
ChatGPT Will Sell Ads in the Seconds Its Images Take to Draw
OpenAI is putting picture ads in the wait while ChatGPT makes an image, a way for a costly free feature to pay for itself that will likely spread to other slow AI tasks.
· 2 min read
Job Interviews That Allow AI Grade Judgment, Not Answers
Interviews that let candidates use AI move the grading from the finished answer to the reasoning behind it, as HackerRank's AI interviewer now does for developers.
· 2 min read
Crisis Plans Cover the Disaster Someone Imagined
Crisis plans cover the disaster someone imagined, so the better habit is to rehearse losing a real dependency, and AI helps in that rehearsal only when it is used with structure.
· 3 min read
AI Answers Are Only Trusted When Someone Defines Revenue
Staff stop using AI tools whose numbers don't match the ones they know, and the old fix still works: write down once what words like revenue mean.
· 2 min read
The Work AI Takes Over Is How Reviewers Learn to Spot Mistakes
When AI does the junior work, the people who later check its output lose the practice that taught them to spot errors, and aviation shows the fix is deliberate practice.
· 2 min read
AI Hiring Is Repeating Programming's Mistake With Women
When a new field starts paying well, whoever controls the way in decides who benefits; programming pushed women out that way, and AI hiring, where women get one in four new jobs, is heading the same way.
· 2 min read
Finance AI Pays Off in Judgment Calls, Not Cost Cuts
New KPMG and PwC research finds finance teams get the biggest AI payoff on judgment heavy work like forecasting and risk spotting, not on cutting costs, and the gap between AI winners and laggards comes down to governance and workforce skills rather than the technology itself.
· 2 min read
Researcher Used AI to Crack Georgia's Secret Ballots
A Princeton researcher used a normal $20 AI chatbot subscription to unmask how more than 1.5 million Georgia voters cast their secret ballots, exposing a known software flaw the state has refused to fix for years.
· 2 min read
Apple Tightens macOS File Access After AI Privacy Scares
Apple is adding stricter consent rules to macOS's Full Disk Access setting after Meta's Muse AI agent was accused of reading a journalist's private messages, a sign that AI agents on personal computers have outgrown the old permission system.
· 2 min read