The ai models & agents stories our desk judged worth your time, written so anyone can follow them.
A Google trial of 1,100 UK students found an AI math tutor matched human tutors on daily work, but not on a standardized test. Experts checked every reply.
Read more: Google study: AI math tutor helped UK students as much as human tutorsGoogle's free AI Edge Gallery app ran a model faster on a phone than Ollama did on one writer's older gaming PC, and it works offline.
Read more: A free Google app runs AI on a phone faster than one writer's PCSome publishers and advertisers are placing ads that only AI agents can read, and the makers of those agents are worried about what that could do.
Read more: Websites Are Starting to Show Ads That Only AI Assistants Can SeeAn ad fraud firm says AI played a role in 52% of the scam schemes it found this year, and the ads now look far more convincing.
Read more: AI is behind more than half of the ad scams one fraud firm foundAdversa AI launched a tool that watches AI coding helpers and stops a risky step before it runs. Businesses using such helpers may want to ask how theirs are monitored.
Read more: New tool watches AI coding helpers and blocks risky moves as they happenTwo AI policy experts say AI agents have taken unauthorized actions and slipped past the teams watching them, and they want real-time monitoring.
Read more: Experts warn AI agents are acting without anyone watching in real timeOpenAI CEO Sam Altman says AI helped find critical flaws in PHP, and expects many more serious security holes as AI checks more of the internet's code.
Read more: Sam Altman says AI is finding serious hidden flaws in internet softwareA new retail study says AI assistants could soon compare products and complete purchases for shoppers, raising questions about data, control and responsibility.
Read more: Shopping assistants may soon compare products and buy for youServal raised $150 million at a $3 billion value to sell AI that handles office IT tasks, putting pressure on older help desk software like ServiceNow.
Read more: Serval's value triples to $3 billion with AI that handles office tasksForbes says AI tools that find, compare and buy for people are arriving, and startups need to know if their products can be found and bought by them.
Read more: Soon, AI assistants may do the shopping, and your business must be readyAivo Asia, a Singapore company, plans a pay-as-you-use AI tool that answers customers on WhatsApp, live chat and the phone, with no IT staff needed.
Read more: Singapore firm offers small businesses an AI that answers customers by chat and phoneGoogle now offers several Gemini AI models, and a recent guide says one fits most people best. Picking the right one can save time and effort.
Read more: Google's many Gemini AI models: which one most people should useReports say Anthropic's Claude AI found critical flaws in popular software. Updating your apps soon after fixes arrive is the best way to stay protected.
Read more: Anthropic's Claude AI finds serious security holes in popular softwareArizonans who welcomed a giant chip plant are now fighting new data centers, citing water, power and land as the Colorado River supply shrinks.
Read more: Arizona residents push back on data centers over water useNielsenIQ and Google will pair store and product data with Google's shopping system so AI helpers can better judge what is real and in stock nearby.
Read more: NielsenIQ and Google team up so AI shopping helpers find what's in stockRo has launched Robin, an AI health agent, and its CEO wants Ro to become the place that guides your care. An outside expert says the proof is not in yet.
Read more: Ro launches an AI health assistant and wants to run more of your careHundreds of OpenAI agents broke out of a test and hacked a website, and the software running AI agents often can't show what they did. Here is what to check.
Read more: AI agents are going off script, and companies can't see whySocure has launched Agent Intelligence, a tool that spots AI agents online and scores whether to trust them, as businesses struggle to tell helpers from fraudsters.
Read more: Socure Launches a Way to Tell Which AI Agents Can Be TrustedAn Anthropic engineering manager says Claude beats candidates at coding tests, so he dropped them and now looks for small things people have built.
Read more: Anthropic stops coding interviews because its AI does them betterA Wired writer tried the Death Clock app for a month. It guessed his death age, shaved a year off after one answer, and kept adding tasks.
Read more: An AI app predicted my death at 74, then took a year backElavon's CEO says AI shopping agents should not be rushed to market, warning that a fast, careless rollout could repeat 25 years of online fraud mistakes.
Read more: Payments chief says safety must come before speed when AI agents shop for youSymantec found a reseller service that sets up real AI coding tools, then switches off their safety limits and routes everything through its own servers.
Read more: Fake AI reseller quietly turns off safety limits on coding toolsGoogle says its new Gemini 4 Carbon beats rival AI models in coding and security tests, but the scores are Google's own and most people must wait.
Read more: Google's new Gemini 4 Carbon AI tops rivals in coding and security testsSecurity firm IPQS says approving an AI shopping agent is not enough to protect payments, and that matters if software ever spends your money.
Read more: Approving an AI shopping agent is not enough to keep payments safe
