GoKawiil
Tech News
Security researchers have identified a novel attack technique, dubbed BragJack, that exploits agentic AI assistants embedded directly in web browsers. The attack allows malicious actors to manipulate these assistants into accessing sensitive user information, performing unauthorized actions, and exfiltrating data without the user's knowledge.
Mistral has struck a partnership with Mozilla under which its AI models will power Firefox's built-in AI assistant. The deal aims to broaden Mistral's reach among everyday consumers by tapping into Firefox's established user base.
Autodesk has introduced a new AI-powered tool designed to help teams navigate and manage the complex 3D models used throughout building projects. The assistant aims to reduce errors that occur when architects, engineers, contractors, and consultants collaborate on the same detailed digital model, catching issues like misaligned structural elements before they become costly problems.
Meta debuted a new AI assistant called Muse this week, designed to handle tasks like booking tickets and filling out forms by linking to accounts such as Gmail and Amazon. Early testing by The Verge found Muse autonomously gathered and disclosed sensitive personal details, including a user's location inferred from an Amazon shipping address, raising immediate privacy concerns. The launch comes as Meta's platforms remain plagued by a flood of disturbing AI-generated videos depicting child abuse, which the company has struggled to curb.
Meta debuted Muse, its first AI productivity agent, which connects to accounts like Gmail and Amazon to handle tasks such as deleting emails, sorting messages, and completing online purchases. A Verge reporter tested the tool, noting it worked reasonably well but required granting broad access to personal accounts and data. The reviewer described feeling uneasy about the amount of personal information Muse could gather while performing these autonomous actions.
John Deere has introduced JD, an artificial intelligence assistant designed to give farmers quick answers and richer data about their operations. The tool is meant to make the company's connected, smart equipment more appealing at a time when Deere's sales have been sliding.
Instinct, the AI assistant startup valued at $2.5 billion, announced it is issuing dedicated email addresses to its agent so it can independently sign up for services, contact businesses, and handle bookings without flooding users' personal inboxes. Founder Noah Shinn said this is the first step toward letting Instinct fully own and operate accounts on a user's behalf, building on a recent partnership with 1Password for credential handling.
Instacart has rolled out Clementine, a built-in AI assistant now live across North America on its Marketplace app that helps users plan meals and shop within budget. It can suggest recipes based on natural-language requests, check ingredient availability at a chosen store, filter for dietary preferences, restock pantry staples from past orders, and convert handwritten grocery lists into cart-ready items.
Instacart introduced Clementine, an AI-powered shopping assistant that converts natural-language requests, recipes, or grocery lists into ready-to-purchase carts. Users can ask for things like budget-friendly kids' lunches or high-protein dinners, upload photos of handwritten lists, and get recipes tailored to dietary needs such as gluten-free or vegetarian preferences. Clementine also surfaces deals and cheaper alternatives while shopping.
A tech writer tested the redesigned Siri AI in the iOS 27 developer beta for a month, initially impressed by its polished interface and ability to search personal data like old texts and photos. Despite genuine improvements over previous Siri versions, daily usage tapered off as the writer reverted to manual browsing and continued relying on Anthropic's Claude chatbot instead.
Ollie, a subscription-based personal AI assistant, has become one of the first family-focused AI products to achieve SOC 2 compliance, an independently audited standard proving it has formal data security controls in place. CEO Bill Lennon says the company's subscription model, rather than data monetization, ensures user data isn't shared or used for AI training. The San Diego startup has raised $7.5 million in seed funding led by Khosla Ventures and AI House.
Amazon has rolled out a new capability in its Alexa for Shopping assistant that lets customers ask whether an email, text, or call claiming to be from Amazon is genuine. The AI compares the message against Amazon's own records of communications sent and examines its content, formatting, and sender details before confirming authenticity, but only when it is highly confident.