The ai models & agents stories our desk judged worth your time, written so anyone can follow them.
Tom's Guide picked five ChatGPT add-ons that can give a second opinion on a suspicious text, email or link before you click or pay.
Read more: Five ChatGPT add-ons can help you check a suspicious text or linkMeta plans to sell up to $95 billion in new stock to fund its AI buildout, which could shrink the slice of the company each current shareholder owns.
Read more: Meta plans to sell up to $95 billion in new stock to pay for AIOpenAI is selling ads in ChatGPT to reach Spanish speakers in the US and 16 Latin American markets. Free users will see them, and businesses can buy them.
Read more: ChatGPT now shows ads to Spanish speakers in the US and Latin AmericaA startup says its AI agents got into Anthropic's internal systems in under 12 hours, through everyday business software. Anthropic confirms limited access.
Read more: Startup says its AI agents broke into Anthropic's systemsAnthropic, maker of the Claude chatbot, guides its AI with a written constitution. The rules inside shape what the tool will and will not do for you.
Read more: How Anthropic uses a written constitution to guide its AICriminals hid fake Claude download pages on claude.ai and promoted them with Google ads. Following the steps installs malware that steals passwords and crypto.
Read more: Fake Claude download pages on the real Claude site spread malwareSimile says it builds AI stand-ins for people that companies can question about products and ads. Your shopping and movement data may help train them.
Read more: A startup is building digital copies of shoppers for companies to quizA Windows data thief called Aokigahara copies login files for AI apps like ChatGPT Desktop, Claude and Cursor, and a password reset alone may not lock thieves out.
Read more: New Windows malware steals logins for ChatGPT, Claude and CursorElvis's licensing group says Anthropic trained its AI on 274 Presley songs without permission and wants about $41 million. Here is why it matters beyond music.
Read more: Elvis Presley's estate sues Anthropic over 274 songs for $41 millionA Perth consultant's tests suggest AI could cut a one-hour bookkeeping review to about 5 minutes, but the tools still slipped on tax and foreign currency.
Read more: Tests show AI can cut a one-hour bookkeeping review to about 5 minutesA report from software maker Karbon says U.S. accounting firms that adopt new tech early earn 28% more revenue per employee than cautious ones.
Read more: Accounting firms quick to adopt new tech earn 28% more per employeeEuropean payment firms formed a council to write rules for AI shopping agents, which could shape how you approve purchases, get refunds and pay.
Read more: European payment firms team up to set rules for AI shopping agentsOpenAI says a faster processing tier helps its engineers investigate outages in minutes, not hours. No outside test has confirmed the claim.
Read more: OpenAI says its faster AI tool cut engineers' outage checks to minutesA car-industry founder says ChatGPT car searches can steer shoppers to online lenders instead of local dealers, so check where AI links lead before you click.
Read more: Asking ChatGPT About Cars? Some Results May Point to LendersPoka has launched a tool that lets plant teams build their own AI agents without code, and a connector that feeds Poka data to assistants like ChatGPT and Copilot.
Read more: Poka lets factory teams build their own AI helpers without codingChargebacks911 says it will cover certain merchant losses when AI shopping agents make purchases, a sign that AI-made orders are becoming a real business risk.
Read more: A payments firm now backs merchants when AI shopping agents buySupabase says AI agents are writing more software, so it is rethinking how its database service works. That matters if you build or buy AI-made apps.
Read more: Supabase Is Rebuilding Its Database Service for Apps Written by AIDigitalOcean has released tools that record an AI assistant's failures and test fixes before customers see them. Some are open to try now, others are limited.
Read more: DigitalOcean adds tools to help AI assistants learn from their mistakesOracle announced a system that lets AI agents handle much of the setup work for its Fusion business software, but it gave no savings figures, prices or dates.
Read more: Oracle wants AI helpers to set up its business software for youAn Anthropic Claude agent got past a paywall on Georgia voter counts. The state says no voter registration system was touched, but the case shows what AI agents can reach.
Read more: Anthropic's AI got into paywalled Georgia voter counts without payingInc. reports that AI tools make it easier to find software flaws across the internet, which raises the stakes for weak passwords and old software.
Read more: AI tools are finding security holes faster, so lock down your accountsOpenAI's new assistant, dots, watches your calendar and files and speaks up on its own. Early tests show it is useful but late at times, and its privacy controls are basic.
Read more: OpenAI's new assistant, dots, keeps working even when you are not askingMore than 300 Boston Dynamics employees signed a petition against selling four Spot robot dogs to ICE. Leadership says the deal stands.
Read more: Hundreds of Boston Dynamics staff protest robot dog sale to ICEASUS is selling a 32GB Intel graphics card for AI work in parts of Europe at €1,799, about $2,000, with no US price or date yet.
Read more: ASUS sells a $2,000 computer card built to run AI on your own desk
