NerdNews

A new report by Common Sense Media's Youth AI Safety Institute found that Google's AI Overview and AI Mode are not safe for kids and teens due to numerous safety issues, including failing to recognize crisis situations and providing inaccurate information. The report recommends that schools and families disable these features and consider alternative search engines.

Stripe introduced a benchmark suite to evaluate AI agents' ability to build real Stripe integrations. The benchmark shows AI agents can build integrations but struggle with validation, particularly in cross-system validation and state tracking.

Microsoft released a record number of security patches for Windows, Office, and other products, citing the use of AI to aid in discovering code vulnerabilities. At least two of the vulnerabilities are zero-days, and one bug affects Windows Server, allowing hackers to escalate privileges.

Apple's generative AI, Apple Intelligence, is coming to China after receiving approval from the Cyberspace Administration of China. The service will integrate Alibaba's Qwen AI model into Apple's operating systems, including iOS, iPadOS, macOS, and visionOS.

Microsoft has released a record 570 CVE updates, including 3 zero-day vulnerabilities, as the use of AI in finding new flaws increases the volume of security updates. Experts warn that this is the new normal and organizations need to adapt their patching strategies to keep up.

Five malicious versions of AsyncAPI packages were published to the Node Package Manager (npm) in a supply-chain attack, delivering a remote access trojan with info-stealing capabilities. The packages had a cumulative weekly download count of over 2.25 million. Developers should pin to known-good files, regenerate lock files, remove the hidden payload, terminate malicious processes, and rotate credentials on impacted systems.

Researchers at Intruder built a pipeline using LLMs and code scanning frameworks to find novel vulnerabilities. The pipeline discovered a zero-day SQL injection vulnerability in a WordPress plugin with over 300,000 users. The tool uses a technique called program slicing to focus on relevant code segments, increasing accuracy and reducing false positives.

Google is celebrating the 25th anniversary of its image search by revamping the feature with more images and AI-powered suggestions. The new interface will display a gallery of images based on users' interests, and also allow for AI-generated images in search results.

Gwen Shapira discusses how teams are scaling AI features using PostgreSQL for mission-critical apps, leveraging Postgres's multi-modal capabilities to deliver deterministic and semantic context to LLMs.

OpenAI has launched its first hardware product, a $230 mini keyboard called Codex Micro, designed for its AI coding agent Codex. The device is a programmable macro pad with 13 mechanical switches, a touch sensor, and a joystick, allowing users to customize their workflow and interact with AI agents.

OpenAI has launched the Codex Micro, a physical keypad for controlling its agentic coding feature Codex. The keypad, created in collaboration with Work Louder, features six frosted keys, a dial, and a joystick, and is customizable. It is available to order for $230.

Whatnot acquired Shaped, a machine learning company, to improve its live shopping recommendations. The deal aims to strengthen Whatnot's discovery and personalization capabilities, helping shoppers find products in real-time as inventory and auctions change.

India has unveiled a ₹625 billion program to incentivize smartphone manufacturing, aiming to reduce dependence on China. The program will run for five years, offering rewards to manufacturers based on eligible sales, with additional incentives for sourcing components locally. This move is expected to create 60,000 direct jobs and generate ₹39 trillion in mobile phone production. India aims to account for 35-40% of global mobile phone production, challenging China's dominance.

Reelful is an iOS app that uses AI to automatically turn photos and video clips into polished short-form videos for social media. It works by getting users to enter a prompt, create a voice clone, and select photos and videos from their camera roll. The app then plans the video, writes the script, adds an AI voiceover, and assembles the final edit.

Emergent, an Indian AI coding startup, has raised $130 million in a Series C funding round, valuing the company at $1.5 billion. The startup provides a platform for non-technical users to build and deploy applications using AI. Emergent plans to use the fresh capital to accelerate product development and research, and expand its go-to-market operations.

Google will allow third-party app stores to access its app catalog starting July 22, complying with a court order from its battle with Epic Games. Third-party stores will have to pay a service fee and meet certain requirements. App downloads will still be completed through Google Play, with Google's service fees applying.

OpenAI's first device will be a rechargeable, humanlike speaker powered by an advanced version of the GPT-Live model. It will have mechanical elements to create an illusion of being alive, a camera, and other sensors to gather context for personalized responses. The speaker is expected to be released in 2027, but its launch may be delayed due to a lawsuit from Apple over trade secret theft.

Google and Epic have withdrawn their attempt to settle a lawsuit, allowing third-party Android app stores to be carried within the Google Play store. This change is set to happen on July 22nd, with Google informing US app developers that their apps will be provided to third-party stores unless they opt out.

Rime, a voice AI startup, has raised $24 million in Series A funding to help enterprises field customer calls. The company uses its own conversational data to train voice models, reducing customization load for clients. Rime's approach focuses on pronunciation of brand entities and industry-specific terms, and it employs a phoneme-based architecture to adapt to different pronunciations.

Every's business operations team uses AI models like Fable, Codex, and Fin to automate tasks, improve efficiency, and enhance customer support. The team's workflow involves 'surfing the models' to hand over complex work to AI agents, allowing them to focus on high-priority tasks. The article also discusses Granola's plan to turn meetings into context that any agent can use and the emergence of new roles in the AI era.