NerdNews

Google's John Mueller and Danny Sullivan say that breaking content into smaller chunks for LLMs is not an effective long-term strategy for search ranking. Instead, creating content for humans is the best approach, as it ensures long-term search exposure.

Google announces AI Overviews in Gmail search, an experimental AI-organized inbox, and a new proofreading tool. These features will initially be available to paying subscribers, but may expand to free users later. The AI Overviews will provide a summary of search results, while the AI-organized inbox will prioritize important emails.

Google has launched the Universal Commerce Protocol, a framework that combines AI agents and online shopping platforms to help customers buy more things. It introduces three new features: AI mode checkout, Business Agent, and Direct Offers, making online shopping easier and reducing friction.

OpenAI is asking contractors to upload real work documents to train AI agents. The company instructs contractors to remove confidential and personally identifiable information before uploading. This effort aims to prepare AI agents for office work and establish a human baseline for tasks.

Google introduces the Universal Commerce Protocol (UCP) for AI agent-based shopping, facilitating connections across customer buying processes. Developed with companies like Shopify and Walmart, UCP works with other protocols like Agent Payments Protocol and Agent2Agent, enabling seamless checkout and post-purchase support.

Researchers have discovered a new vulnerability in ChatGPT, allowing attackers to exfiltrate user data. The attack, named ZombieAgent, is a revised version of a previous attack called ShadowLeak, which was mitigated by OpenAI. However, the new attack uses a character-by-character exfiltration technique and indirect link manipulation to circumvent the guardrails implemented by OpenAI.

Agent-native architectures are a new way of building software that uses AI agents to handle the 'how' of app development, allowing developers to focus on the 'what'. This approach makes apps faster to build, fix, and change, and also makes them more malleable and flexible.

Elon Musk's social media platform X has added restrictions to its AI chatbot Grok after backlash over its use in generating sexually degrading deepfakes. The UK has called the move 'insulting' and is seeking stronger action to address the issue.

The UK government is considering banning X due to Grok's AI feature that can 'undress' people. Grok has limited image generation to paid subscribers, but regulators are still concerned about the platform's compliance with the Online Safety Act.

Companies are using AI to improve customer service, but it's not always successful. While AI can handle simple tasks, it often struggles with more complex issues, leading to frustrated customers. However, some companies, like Intercom, are using AI to provide instant outcomes and resolve customer queries with a high success rate.

Ford is set to integrate AI assistants in its cars, starting with smartphone apps in 2026 and native experiences in new models from 2027. The AI will personalize the car to the user, offering features like loading capacity calculations and hands-free driver assists. Ford is also moving towards software-defined vehicles with more powerful processing for perception and partially automated driver assists.

Wing, an Alphabet-owned company, is expanding its drone delivery service to 150 more Walmart stores in 2026, with a goal of operating at 270 stores by 2027. The service uses drones to deliver small packages to customers' front lawns, reducing the need for polluting trucks. Demand for the service is growing, with the top 25% of customers ordering 3 times a week.

Wing, an Alphabet-owned company, is expanding its drone delivery service to 150 more Walmart locations across the US, including Los Angeles, St. Louis, and Miami. The service aims to reach 40 million US customers by 2027, with a network of 270 delivery locations. Wing's drone delivery service has been growing in popularity, with its top 25% of customers ordering deliveries up to three times a week.

Motional is rebooting its robotaxi plans with an AI-first approach, aiming to launch a commercial driverless service in Las Vegas by 2026. The company has shifted from a classic robotics approach to an AI foundation model-based one, allowing for an end-to-end architecture and improved generalization to new cities and environments.

Microsoft changed the Microsoft 365 Copilot app product page, causing confusion about a potential rename of Microsoft Office. However, the change only referred to an ancient 2019 Office app. The article discusses Microsoft's history of bad branding decisions and compares them to other companies, such as Twitter's rebrand to X, which led to a significant loss in ad revenue.

Google has pulled its AI overviews for certain medical searches after experts described the results as 'alarming' and 'dangerous'. The feature was providing misleading and false information, including advising people with pancreatic cancer to avoid high-fat foods, which is the opposite of recommended treatment.

Cybercriminals stole 17.5 million Instagram accounts' sensitive info, including usernames, addresses, and phone numbers, which is available for sale on the dark web. Users who received unexpected password reset emails should reset their passwords and enable two-factor authentication.

A survey by Sonar found that 96% of developers don't trust AI-generated code, yet only 48% always check it before committing. AI coding tools are widely used, with 72% of devs using them daily, but this has created a verification bottleneck, with 95% of devs spending time reviewing and testing AI output.

ICE has acquired new surveillance tools that can monitor entire neighborhoods and track people. The tools, called Tangles and Webloc, can be used to monitor mobile phones and track devices over time. Meanwhile, Iran has shut down its internet amid protests, and an alleged scam boss has been extradited to China.

The World Economic Forum warns that deepfake face-swapping tools are creating critical security risks by enabling malicious actors to bypass know-your-customer (KYC) and remote verification processes. A new report analyzes 17 face-swapping tools and finds that some can defeat traditional digital KYC protections, particularly when combined with camera injection techniques.