NerdNews

OpenAI is rolling out GPT-6 to all tiers of ChatGPT, featuring Intelligent UI for interactive user interfaces, enabling quick responses with visuals like graphics, buttons, and charts. The model is trained for better everyday conversations and can compose visual responses within the chatbot.

Nvidia's Halos system provides a full-stack safety solution for physical AI, enabling safer deployments of autonomous mobile robots and humanoid robots. The system includes hardware and software tools, such as the Nvidia IGX Thor computing module and the Halos operating system, to monitor and isolate safety-critical computing workloads. Companies like Agility Robotics and Boston Dynamics are already using Nvidia Halos to improve safety in their robots.

Microsoft unveiled the Surface Laptop Ultra and RTX Spark Dev Box, both featuring Nvidia's RTX Spark SoC. The company also announced updates to Windows 11, including a new Copilot layout, improved Windows Search, and Microsoft Execution Containers for secure AI agent deployment. These changes aim to bring local AI capabilities to Windows users, with a focus on hybrid intelligence and agentic workflows.

Pratik Rasam discusses Spotify's multi-agent AI architecture for advertising, sharing key patterns, domain ownership models, and evaluation strategies for running production-grade multi-agent systems at scale.

Cloudflare announced Clef, a set of open-weight AI models designed to choose between predefined options. The 27B-parameter model can process text, JSON, images, or video and returns a probability for each allowed option. Cloudflare also released Clef-Flash, a smaller 9B multimodal model for latency-sensitive decisions.

Microsoft Teams will soon introduce support for third-party deepfake detection solutions and impersonation protection in Teams meetings. The changes are part of a broader move to boost security for Teams users and will reach general availability in November.

Google's AI Edge Foresight app can transcribe meetings and audio files entirely offline using an on-device model, allowing users to take notes and ask questions without connecting to the cloud.

Natura's $99 smart ring, Interface, combines AI agents and health tracking, allowing users to control devices and access information with a press of a finger. The ring connects with various AI agents and apps, tracks health metrics, and has a battery life of 6-12 days.

Uber and Pony.ai are partnering to launch a robotaxi service in London, with testing of Pony.ai's Gen-7 robotaxis to begin in the coming weeks. The service will follow a similar model to their existing partnership in Croatia, with Pony.ai supplying the autonomous driving system and Uber providing access to its ride-hailing network.

Vesta, an AI-native software startup, has raised $30 million to automate mortgage lending processes using AI agents, reducing time and costs for lenders. The company's AI agents can perform tasks and track workflows, and some lenders are even using them to make underwriting decisions.

Amazon has announced several new devices in October 2026, including Alexa tablets, Kindles, a Ring smart door lock, and an updated Fire TV Stick 4K. The Alexa tablets come in three sizes and run on Android 16 with access to the Google Play Store. The new Kindles include a standard model, a Paperwhite, and a Colorsoft, all with improved displays and features. The Ring smart door lock includes a hand crank for emergency power and a fingerprint reader. The updated Fire TV Stick 4K is faster and thinner than previous models.

Mistral's new 1 trillion-parameter model, Le Chonk, is a freely available AI model that can compete with top-end models from the US and China. It's optimized for coding, cyberdefense, and specific tasks in manufacturing, finance, and electrical engineering. Le Chonk is considered the most capable open-weight model developed outside of China and is 'very, very close' to some proprietary models.

AI hooks provide deterministic controls for AI coding assistants, allowing security teams to enforce rules at the system level. GitGuardian AI Hooks scan prompts, tool calls, and outputs for secrets, stopping them before they spread. Supported assistants include Claude Code, Cursor, Codex, and more.

Three safety researchers fired by OpenAI have published an open letter denying misconduct claims and warning of a chilling effect on the company's culture. They allege that their dismissal was not for raising safety concerns, but rather for collaborating with external safety evaluators, which they believe is essential for AI safety work.

Waymo, a leading autonomous vehicle technology company, has closed a $5 billion loan from prominent lenders including PIMCO, Blackstone, and Sixth Street. The debt financing will fuel the company's commercial expansion of its robotaxi services in existing cities and new markets in the US, Europe, and Japan.

Amazon's new Kids Tablets come in 8.7-inch and 11-inch sizes, with colorful cases, kid-appropriate software, and a year of Amazon Kids+ subscription. They feature an all-aluminum design, 8-core processor, and up to 14 hours of video playback. The tablets will be available on October 14, starting at $230.

Amazon's latest Alexa tablets will run a custom version of Android and have full access to the Google Play store, offering users a wider range of apps and more freedom to choose. This is a departure from previous Amazon tablets, which were limited to the Amazon Appstore. The new tablets will be available in several countries, including the US, UK, and Canada, starting October 14 and 19.

Cryptographers are debating whether AI could break Bitcoin's security in months, with some experts warning of a potential 'bunker mode' and others dismissing the idea as 'fear mongering'. The debate centers around the possibility of AI-driven mathematical breakthroughs cracking the digital signature scheme used by Bitcoin, specifically the ECDSA signatures. Experts like Justin Drake are warning of a potential break, while others like Yehuda Lindell argue that there is no evidence to support this claim.

Anthropic has launched a new service called OSS Scanner, which offers free AI security scans for open-source projects. The service uses Anthropic's strongest models, including Mythos, to generate vulnerability reports. While the reports are generated without human review, they can help open-source projects identify potential security issues sooner.

A solo developer used AI to recreate Adobe's popular tools, including Photoshop, Premiere, and Illustrator, and gave them away for free. The developer, Brandon Thomas, was inspired by Adobe's cancellation policies and used Rust to build the clones. However, the move raises questions about the future of software and creativity.