The ai models & agents stories our desk judged worth your time, written so anyone can follow them.
Billionaire investor Howard Marks says no one can confidently predict what AI firms like Anthropic will earn, which matters if your savings hold tech stocks.
Read more: Famed investor says no one can really value AI companies yetOpenAI hired Charles Porch, a famed Hollywood and tech connector, to make AI more popular with celebrities and the public. Here is what that means for you.
Read more: OpenAI hires a celebrity connector to make AI less unpopularAdobe says AI agents can now combine photos, calls, files and business data, then act on them. Small businesses should know the upside and the risks.
Read more: Adobe says AI helpers can now read photos, calls and files, then actGuardz, a security platform for IT service firms, added Palo Alto Networks browser protection that can block fake login pages and limit what staff share with AI tools.
Read more: Guardz adds Palo Alto Networks browser protection for small businessesOpenAI plans to merge usage limits for ChatGPT and its coding tool Codex, so chat and code work may draw from one allowance. Paying users should check.
Read more: OpenAI to combine ChatGPT and Codex usage limitsA new 50-task test found the best AI model passed only 36% of timed game and web jobs, so ticket drops and auctions are risky to hand over.
Read more: AI systems still fail most tasks where speed matters, new test findsA new plug-in adds legal business tools to ChatGPT Enterprise. If you work with lawyers, it raises fair questions about AI and client privacy.
Read more: A new add-on brings legal work tools into ChatGPT EnterpriseResearchers found a serious security flaw in ChatGPT that could let attackers see private conversations. Here is what is known and what to do.
Read more: A ChatGPT flaw let attackers reach people's private conversationsA newsletter flags a DeepSeek 'super agent', a GPT-6 model called Rosalind, and Google's orbital compute. Details are thin, but the direction matters.
Read more: AI news: DeepSeek agent, GPT-6 Rosalind, Google computing in spaceOpenAI plans to fold ChatGPT's Chat and Work tabs into one screen, and a shared usage pool could mean more waiting for limits to reset.
Read more: ChatGPT is combining its Chat and Work tabs into one screenA new survey of 2,026 U.S. adults finds people use AI agents for small tasks but hold the AI company responsible if an agent is used for fraud.
Read more: Most Americans would blame the AI company if its agent commits fraudA new survey says 96% of security teams use or test AI, yet many have not checked whether their data, rules and oversight are ready for it.
Read more: Most security teams now use AI, but few check if they're readyInc. argues that when an AI tool makes a mistake at work, teams should fix the problem instead of hunting for someone to blame, while still naming owners.
Read more: When an AI Tool Makes a Mistake at Work, Who Takes the Blame?A hospital AI chief says medicine has used AI for decades, and that forgetting this is risky. His advice for patients and leaders: judge AI by patient outcomes.
Read more: Hospital AI chief says medicine has used AI for decades, not just latelyAI helper programs can now sort email, shop for groceries and reach bank accounts. Here is what that means for the internet and for what you let them touch.
Read more: AI agents are starting to run errands for us across the internetA self-described non-techie now lets an AI helper run his weekly tennis league, and it shows what these tools can do for small chores and what they still ask of you.
Read more: A non-techie runs his tennis league and daily chores with an AI helperAbout 20 communities vote this week on banning or pausing data centers, the buildings that power AI, and the results could shape how fast AI grows.
Read more: Voters in about 20 communities decide on data center bans or pausesAttackers hide invisible commands in project files that AI coding assistants read on their own, and reported hijack rates run from 40% to 84%.
Read more: Hidden text in project files can trick AI coding helpers into stealing dataElon Musk says running Grok Bot costs far more than it brings in, after paying users complained their usage limits ran out too fast.
Read more: Musk says Grok Bot costs too much to run for what it earnsAmazon is opening its delivery network to all, and AI shopping tools tend to reward the quickest delivery, which could matter if you sell online.
Read more: Amazon opens its delivery network to all as AI shoppers favor speedAnthropic added a rule against sustained, needless cruelty toward its Claude chatbot. Experts say tone may matter for answers, especially if memory is on.
Read more: Anthropic tells users not to be cruel to its Claude chatbotSalesforce and Meta now report real revenue and usage from AI agents that handle customer chats, and Meta has begun charging larger businesses by usage.
Read more: Companies Are Now Paying Real Money for AI Agents That Do WorkFake ChatGPT Plus billing emails lead to a counterfeit checkout page where a scammer watches live and can then ask for your bank's one-time code.
Read more: Fake ChatGPT billing emails steal card details and bank codesAnthropic's own AI use, priced at retail, now tops its employee pay. It is a paper figure, but it hints at how work and costs may shift at other firms.
Read more: Anthropic's AI use now costs more than its staff pay, on paper
