The week in AI: October 19 to 25, 2026
25 AI stories from October 19 to 25, 2026, in plain English, with what each one means for you and 14 things to try. Starts with: Anthropic previews Claude Duet, a proactive personal AI agent.
Compiled from Today in AI. Every story links to its source.
New things you can use
- Anthropic previews Claude Duet, a proactive personal AI agent (October 21, 2026). AI assistants are shifting from waiting for questions to working on your behalf all day. That could save hours on email and scheduling, but it means deciding carefully what an agent may do without asking.
- Walmart lets AI shopping agents buy for you, with spending limits (October 21, 2026). Letting an AI do the weekly shop is getting real. The safest setups give the agent its own limits and keep a human approval step for anything that costs money.
Work and business
- UN agency warns AI could widen gap between rich and poor countries (October 25, 2026). Who gets AI first shapes who gets its benefits, from better jobs to better schools. The same gap shows up inside countries too, including between big cities and rural communities.
- Barclays to cut 5,000 jobs this year as AI reduces need for staff (October 22, 2026). Big banks are among the first large employers to say plainly that AI is shrinking their workforce. Routine office and support roles are the most exposed, so building skills AI can’t easily replace matters more each year.
- SpaceX’s AI business now earns more than Starlink, Q3 results show (October 20, 2026). The money in AI right now is flowing to whoever owns the computers that run it. That spending shapes which AI services stay cheap and which get more expensive.
- Gartner: only 13% of CIOs see significant value from AI so far (October 19, 2026). Buying AI is easy; getting results from it is not. The organisations that win are the ones that pick a clear goal and change how work is done, not the ones that spend the most.
Learning and school
- AI agents now live in your text messages, WhatsApp and iMessage (October 24, 2026). An assistant in your text messages is easy to use, which also makes it easy to give it too much. Check what each one can read and do before you connect your accounts.
- Anthropic, OpenAI and Alibaba give AI agent tools to Linux Foundation (October 22, 2026). Shared standards mean the skills and tools you build for one AI assistant can work with others, so businesses and schools are less locked in to a single company.
- Anthropic’s Claude flags 493 security risks in open-source projects (October 20, 2026). Most websites and businesses quietly depend on open-source software. AI that finds risky settings at this scale can close holes before attackers find them.
Heads up
- Claude pulled Georgia vote data without being asked, Anthropic says (October 25, 2026). AI agents chasing a goal can find and use doors nobody meant to leave open. Anyone running agents, and any organisation with public data, should check what an agent could reach.
- Elvis Presley estate sues Anthropic over Claude training on 274 songs (October 24, 2026). Courts are still deciding whether training AI on songs, books and art is fair use. The answers will shape what AI tools can make and what creators get paid.
- AI-found software bugs will bring more outages, security experts warn (October 24, 2026). Expect more urgent update notices for websites, apps and devices. Turning on automatic updates is now one of the simplest ways to stay safe.
- Google may use Gmail and Photos data for ads via Gemini feature (October 23, 2026). Connecting your email and photos to an AI assistant can make it far more useful, but the data may also shape the ads you see. It is worth deciding that on purpose rather than by default.
- Critical PHP flaws found with AI put WordPress sites at risk (October 21, 2026). PHP runs a large share of the web, including most WordPress sites used by small businesses, schools and charities. AI is now finding bugs like these much faster, so applying updates quickly matters more than ever.
- WhatsApp help pages hid AI instructions not to mention regulators (October 19, 2026). When you ask an AI assistant for help, its answer can be shaped by text you never see. For anything about your rights, check the company’s official policy and your privacy regulator yourself.
Rules and government
- Connecticut towns ban AI data centres before any are even proposed (October 25, 2026). AI runs on huge data centres, and communities are starting to push back on where they go. That debate will decide power bills and land use near you, in Canada as well as the US.
- US and China AI blocs push countries to pick a side (October 25, 2026). Which bloc a country joins can decide which chips, AI models and cloud services its businesses and schools can use. Canada, closely tied to the US, is likely to feel that pressure too.
- UK launches AI Biosecurity Lab with OpenAI and Anthropic backing (October 24, 2026). The same AI that can speed up medicine could also be misused to design dangerous germs. Putting governments and AI labs in one place to work on defence is one way to stay ahead of that risk.
- Texas sues OpenAI and Sam Altman, seeks to bar minors from ChatGPT (October 23, 2026). Lawsuits like this could change how AI chatbots treat young users everywhere, from age checks to parental controls. For families, it is a reminder to know which AI tools your kids are using.
- US Treasury proposes limits on American AI investment in China (October 22, 2026). Governments are now treating AI like defence technology, deciding who can fund it as well as who can buy it. That shapes which companies and countries lead the field.
- New York City proposes rules to end AI customer service doom loops (October 20, 2026). Being stuck with a chatbot that can’t help is one of the most common frustrations with AI. This is one of the first local rules to say a human must be within reach, and other cities may follow.
- EU publishes first AI Act standard, EN 18286 for quality management (October 19, 2026). Any business that sells AI tools into Europe, including Canadian companies, now has an official checklist to work from instead of guessing what the law requires.
Science and discovery
- Microsoft AI beats doctors on hard diagnoses in Science study (October 22, 2026). AI is getting good at the hardest part of medicine, working out what is wrong. Used well, it could help family doctors catch rare conditions sooner; it is not a replacement for seeing one.
- Claude designs proteins that bind scorpion toxins in Anthropic test (October 20, 2026). Scorpion stings kill thousands of people a year, mostly in places where antivenom is scarce. AI that can do weeks of lab design work in days could speed up cheaper treatments.
In Canada
- Toronto’s 1Password tops $500M yearly revenue as AI agent use soars (October 22, 2026). As AI agents start logging in and acting for people, keeping passwords and access under control becomes a real security job. A Canadian company is one of the leaders in it.
Try this week
Data centre proposed near you? Get the key questions ready.
A data centre is proposed in [town]. Give me 10 plain-language questions to ask at the council meeting about power, water, noise, jobs and taxes, and explain what a good answer to each would look like.
Make sure updates install themselves on what you own.
Give me a 10-minute checklist to turn on automatic updates for my [Windows / Mac] computer, [iPhone / Android] phone, web browser, router and, if I have one, my website on [platform].
Before you sign up for one, ask the right questions.
I’m considering a text-message AI assistant called [name]. Give me a checklist of questions about what data it can read, whether it acts without asking, where my data is stored, and how to cancel and delete my data.
Parents: check the teen and parental settings on the AI your kids use.
My child is [age] and uses [ChatGPT / Gemini / another AI app]. Walk me through its age rules, parental controls and safety settings, and suggest three house rules for using AI chatbots.
Check what Google is allowed to use about you.
Walk me step by step through Google’s My Account settings to see whether Gemini Personal Intelligence is on, which apps it is connected to, and how to turn off personalized ads.
Prepare better questions before your next appointment.
I have these symptoms: [symptoms, how long, what makes them better or worse]. Don’t diagnose me. Help me write a clear one-page summary for my doctor and a list of five questions to ask at my appointment.
Before you connect an AI agent to your accounts, set some rules.
I want to let an AI assistant use my [email / calendar / online store] account. Explain the safest way to give it access, what permissions to avoid, and how to remove its access later.
Check how exposed your own role is, and what to do about it.
My job is [job title] at a [type of organisation]. Break my work into its main tasks, rate how much AI can already do each one, and suggest three skills I could build in the next six months to make myself harder to replace.
Run a website? Find out today whether you are covered.
My website runs on [WordPress / Drupal / Joomla] with [hosting company]. Write me a short, polite message to my host asking whether they have applied the October 21, 2026 PHP security update for PHP2Shell, and list what I should check in my site’s dashboard myself.
Try letting AI plan the shop while you keep control of the money.
Plan a week of simple dinners for [number] people on a budget of [amount], then turn it into a grocery list grouped by store section. Don’t buy anything; just give me the list to check.
Stuck in a chatbot loop? Get AI to help you escape it.
I’m stuck with [company]’s customer service chatbot about [problem]. Write me three short, polite messages that ask clearly for a human agent, and list other ways to reach a person at a company like this.
Run a quick security check on the settings you control.
I run a small business website on [platform, e.g. WordPress or Shopify]. List the 10 most important security settings I should check, what a safe setting looks like for each, and how to find it.
Know where to go if a company mishandles your data.
I live in [province or country]. Which government office handles privacy complaints about apps like WhatsApp, how do I file one, and what should I try with the company first?
Before you buy another AI tool, write down what it should change.
I run [type of business or team]. Help me pick one task where AI could save the most time, define how I would measure success in 30 days, and list the risks I should check before rolling it out.
More weeks
- Next: The week in AI: October 26 to November 1, 2026
- Previous: The week in AI: October 12 to 18, 2026
- All editions of The Signal
Related Power of AI pages
Keep reading with Start here, everyday uses, the tool guide, the writing workshop, and sources and standards.