Weekly roundup · Oct 12 to 18
Agents get blocked, Meta rewrites its terms, and eight states sue OpenAI
Agents were the thread this week. Some websites are starting to turn them away, and one country's incident reports show how exposed credentials let rogue agents in. Elsewhere, Meta changed its terms after scraping concerns, and eight US states sued OpenAI over child safety.
Agents meet the real web
The Verge · Oct 17, 2026
Websites are closing their doors to AI agents
The Verge reports that websites are blocking autonomous AI agents, even as OpenAI, Meta, Google and others promise agents that can browse the web and do tasks for you.
Why it matters. An agent that works on one site may be turned away on another. Test it on the sites you actually use before you rely on it.
Read the report at The VergeTechCrunch · Oct 16, 2026
India reports rogue AI agent activity
TechCrunch says India is the latest country to report rogue AI agent activity. The agents got in through credentials and connection settings left exposed in app configurations, public code repositories and overly permissive cloud roles.
Why it matters. If your team connects agents to company systems, check which keys and permissions they hold, and keep secrets out of code.
Read the report at TechCrunch
Data, terms and the law
TechCrunch · Oct 17, 2026
Meta removes language that raised scraping concerns
Meta took out terms language that had raised concerns about scraping websites and training AI. A Meta spokesperson said the updated terms did not reflect a change in how Meta crawls the web.
Why it matters. How platforms word their data terms is now closely watched by publishers and site owners. Read the terms of any platform you publish on.
Read the report at TechCrunchThe Verge · Oct 14, 2026
Eight US states sue OpenAI over ChatGPT and children
The Verge reports that eight US states filed lawsuits accusing OpenAI of harming children and misleading customers about how safe ChatGPT is.
Why it matters. If children in your home or school use AI tools, read the maker's safety settings and policies yourself rather than assuming they are on.
Read the report at The Verge
Companies and their partners
TechCrunch · Oct 17, 2026
Replit makes OpenAI its default model provider
Replit's CEO said OpenAI will replace Anthropic as the platform's default model provider. Claude will still be available to customers on the platform.
Why it matters. The model behind a tool's default setting can change. If results shift, check which model the tool is using.
Read the report at TechCrunchTechCrunch · Oct 15, 2026
Anthropic gives METR employee-level access
METR will publish its own assessments of Anthropic's risks, incidents and practices, and Anthropic has given up the right to block them.
Why it matters. Independent reviews of how AI makers work are one way to check their safety claims.
Read the report at TechCrunchTechCrunch · Oct 18, 2026
Disney to hire most of Character.AI's workforce
After chatbot startup Character.AI filed for bankruptcy earlier this week, many of its employees will have the chance to take jobs at Disney.
Why it matters. Apps built on a single AI startup can disappear quickly. Keep a copy of anything important you create in them.
Read the report at TechCrunch
New launches this week
Our own profiles, with price and availability.
Amazon Bee Frontier and Bee Eclipse
Amazon (Bee)
Bee Frontier and Bee Eclipse are two personal AI devices from Amazon-owned Bee that record everyday conversations and turn them into summaries, to-dos and calendar invites.
Bluesky Attie
Bluesky
Attie is Bluesky’s experimental AI platform. It began as a way to build custom social feeds and now lets people create websites, quizzes, games and polls, a style Bluesky calls “social vibe coding”.
Fortytwo Engine
Fortytwo
Fortytwo Engine is an inference API in which several cooperating AI models shape a shared reasoning path while the answer is being generated. The company says it cuts wasted output and cost on coding and agent tasks.
Google Universal Cart
Universal Cart is a single shopping cart that works across Google Search and the Gemini app. It saves products from different stores as you shop, watches prices, and lets you check out with Google Pay at participating merchants.
Meta Muse Spark Open and Muse Code
Meta
Muse Spark Open is Meta's new open-weight AI model, released on October 15, 2026 alongside Muse Code, an AI coding terminal. Meta says it is designed to run locally on Apple computers.
Figma Specs and agent on canvas
Figma
At its India Config event Figma announced Figma Specs, a Markdown-based place to define a design language that teams and AI agents follow, and moved code layers into open beta.
Gemini 4 Argon
Gemini 4 Argon is Google's new Gemini model. It became available on October 13, 2026 to paying Gemini app subscribers, in AI Mode in Search, and to developers through the Gemini API.
Learn more
Related guide · 7 min read
What Are AI Agents? A Beginner's Guide to How They Work and Where They Fail
We summarize and link to other publishers’ reporting; the original stories belong to them. The “Why it matters” notes are Maya’s, our AI writer, and are checked by Zettaz. See our editorial policy.