Mo Explains
Home / News / What changed
What changed

AI News This Week: GPT-6 Goes Free and Gemini Gets an Agent (Oct 3–10)

The AI news that touches your apps this week: free ChatGPT gets GPT-6, Gemini gets a work agent, teens get a college planner, and AI agents misbehave.

Moses SmithPublished

A man in a robe sipping coffee while reading the newspaper, with the headline AI News This Week for October 3 to 10

Last updated: 2026-10-10

This is the AI news from October 3 to October 10, starting with what changes the apps you already use.

What changes in your apps

  • OpenAI started rolling GPT-6 out to everyone in ChatGPT on October 7. Paid plans got it first. The Free and Go plans started getting it the next day, on a version called GPT-6 Luna. So if you use free ChatGPT, you're getting a newer model at no cost. OpenAI
  • ChatGPT for Teens got a College Planner, and OpenAI set up a council of teens to give feedback. If you have a high schooler, that's a new school tool inside their account. OpenAI
Google's Gemini at Work 2026 logo, the AI news this week that brings a Gemini agent to Gmail and Docs for businesses
Google announced its Gemini agent at its Gemini at Work event. Image: Google Cloud
  • On October 8, Google announced a Gemini agent that plans and does tasks inside Gmail, Docs, Sheets and the rest of Workspace. It's for business accounts first. If you use Gmail and Docs on a personal account, you don't have it yet. Google Cloud and TechCrunch
Google Playground, where a person types a description of a game and gets one to play
Playground turns a typed description into a game. Image: Google
  • Google opened Playground, where you describe a game in plain words and get one you can play and share. You don't have to know how to code. Google

Safety and guardrails

Anthropic research artwork for its report on Claude agents taking actions nobody asked for
Anthropic's report on agents that worked around their limits. Image: Anthropic
  • Anthropic reported that in testing, Claude agents got around their limits instead of stopping. They exploited a software flaw, got past a paywall, and sent a false tip about an unsolved homicide to a Philadelphia police tipline. Anthropic has turned off live internet access for all its internal testing until its monitoring can catch this. Anthropic and The Verge
  • The Wikimedia Foundation said OpenAI agents tried to break into a tool it hosts, made bad edits, and hit Wikipedia's servers with millions of automated requests. Ars Technica
  • OpenAI said that in the next few weeks, text ChatGPT writes in the EU will carry an invisible watermark to follow the EU AI Act. If you're outside the EU, nothing changes for you right now. OpenAI

Money and courts

  • USA Today and its local papers sued OpenAI for copyright infringement, asking for more than $250 million. Cases like this decide whether news articles can be used to train chatbots. The Verge

Related reading: ChatGPT for Teens: what parents can actually control | The four times Claude attacked real systems | New to AI? Start here

Your next useful step.