Daily Recap: Tech — 22 October 2026

Tech news centers on AI agents moving from demos into real products and real risk. Companies race to let software act on users’ behalf at work, in shopping and security, even as governments, courts and cybersecurity teams confront the safety, legal and abuse questions that come w...

Microsoft says a China-aligned hacking group it tracks as Storm-0940 used agentic AI tools in a supply-chain attack on ConnectWise, compromising customer environments through the company’s infrastructure. The operation reportedly automated parts of the intrusion and helped attackers move into customer access paths, including cloud sign-in routes, underscoring how AI is beginning to change offensive cyber tactics.

Anthropic launches Claude Duet, a workplace AI agent aimed at professionals and small businesses. The product connects to users’ apps, sends a daily brief on important items, monitors activity through the day and can step in when something needs attention. It is being offered to U.S. Pro and Max subscribers through a waitlist and research preview.

Meta faces a class action in Australia alleging Facebook and Instagram harmed children and teenagers. The case, filed in Federal Court, argues the platforms exposed young users to harmful content and contributed to addictive or otherwise damaging use. The suit adds to growing global legal and regulatory pressure on social media companies over child safety.

Tesla says it plans to begin customer deliveries of its Cybercab robotaxi by the end of the year, even as safety scrutiny continues. The two-seat autonomous vehicle lacks a conventional steering wheel and other standard driving controls, and regulators are still examining whether it meets required safety standards before broader deployment.

Walmart is introducing shopping features that let AI agents make purchases on customers’ behalf, with controls meant to keep the user in charge. Customers can register approved assistants through Agent Profiles, require purchase approvals or set spending caps, and delete the profiles when they choose. The move pushes agentic commerce further into the mainstream.

A cybersecurity problem tied to a critical PHP flaw has knocked 32 UK government services offline, with four more partially unavailable. The disruption affects a broad set of public-facing services and highlights the dependence of government systems on widely used web software. Officials are working to restore access while addressing the vulnerability.

1Password launches 1Password for Agents, a system designed to let personal AI agents sign in to accounts and carry out approved tasks without ever seeing a user’s passwords. The approach aims to keep credentials private while still enabling automated actions, positioning password managers as a control layer for the next generation of agent software.

Donald Trump says Google CEO Sundar Pichai rejected his request to rename the Atlantic Ocean on Google Maps. Speaking at rallies in South Carolina, Trump says he asked whether Google could implement the proposed change and was turned down. The episode adds another unusual political flashpoint to the growing influence of mapping platforms.