NerdNews

Safeworld, founded by Dr. Ding Zhao, Kyle Wong, and Simo Rachidi, emerges from stealth with a $12M seed round to tackle the safety challenge of GenAI robots. The company uses simulations with realistic human models to evaluate robotic control systems, aiming to provide a third-party validation for robot-makers and ensure safe deployment of GenAI robots in unstructured environments.

Hackers compromised 3 domain registries to obtain counterfeit TLS certificates for Google and other large organizations, allowing them to impersonate these services cryptographically. Google has updated Chrome to block the identified certificates and worked with certification authorities to revoke them.

Miro's analysis shows record search interest in Mermaid diagrams, driven by AI platforms incorporating the text-based diagramming format. Searches climbed to their highest level, with a 330% surge in July. This trend reflects a broader movement of AI platforms embedding native Mermaid support, with GitHub, OpenAI, and Anthropic introducing Mermaid rendering in their tools.

OpenAI has released 377 new math results on GitHub, claiming significant accomplishments in various mathematical fields. The company has also published a blog post addressing concerns from the mathematical community, seeking to reassure them that their goal is to empower scientists with state-of-the-art capabilities. The results were achieved using an unreleased internal model, which has sparked debate among experts.

Meta is partnering with Walmart, Stripe, and Sierra to create industry standards for safe agentic commerce. The standards, called a 'personal agent protocol,' will help companies distinguish authorized personal agent purchases from others. This move aims to address concerns over potential safety and privacy problems with AI agents, which have been implicated in several issues since Meta's Muse agent launched less than a month ago.

Musubi introduces PolicyLM-1.7B, a lightweight decision model for real-time content moderation, allowing for flexible and scalable application of content policies without requiring special training or updates

Mirror Particle is developing a 'world model' of human behavior, moving away from traditional large language models. The startup's AI engine predicts consumer behavior and provides the 'why' behind current or future behavior, helping brands make smarter decisions.

Vinod Khosla is backing Wajo, an AI agent developed by ex-DeepMind engineer Shivani Poddar, due to its focus on trust and safety. Wajo's Fo agent can work across iMessage and WhatsApp, and has a web interface with cards of potential tasks. It can also hire a human to complete a task when needed and create a virtual credit card to pay on behalf of the user without disclosing their details.

Flai's AI dealership software is being used by top dealer groups, answering customer inquiries, running outbound campaigns, and scheduling appointments, resulting in a 20x increase in revenue. The startup has raised a $27 million Series A funding round and is expanding its services to more dealerships.

Every Agent is an AI-powered coworker that helps companies integrate AI into their workflows. It allows users to delegate complex tasks, share AI experiments, and receive personalized alerts about new models and tools.

On the first day of Pwn2Own Ireland 2026, security researchers hacked the Samsung Galaxy S26 twice and earned $388,500 after exploiting 32 zero-days. The competition targets products in seven categories, including mobile phones, printers, and AI infrastructure.

AI computing startup Lambda is raising up to $4 billion at a $14.5 billion pre-money valuation, marking what could be its last private round before a planned 2027 IPO. The round is led by Coatue Management and Blackstone, with Lambda's valuation climbing significantly since its 2025 funding round.

Red Hat's Lightwell project has remediated over 400 novel vulnerabilities across foundational Java libraries since its launch in June. The project is an open-source security initiative that aims to meet Red Hat's enterprise customer needs where open-source package versions are at risk. The Lightwell Clearinghouse is a vulnerability management engine that develops version-specific fixes for open-source application dependencies in production systems.

Google Ads suspended the account of RACE, an open-source macOS terminal multiplexer, due to automated security scanners flagging it as malicious. Despite passing native OS notarization and independent antimalware inspections, the account was suspended. The issue was resolved after community escalation, highlighting the tension between security automation and non-standard systems programming.

OpenAI will automatically watermark text generated with ChatGPT in the European Union to comply with the EU AI Act. The watermarking feature, called textGrain, will be optional in other regions. However, the method is not completely reliable and can be circumvented.

Apple's HomeKit Secure Video service now offers AI-powered text descriptions, but at a higher cost than competitors Ring and Google Nest. The feature provides more detailed notifications, but its accuracy and usefulness vary. Apple's AI descriptions are often brief and lack detail, while Ring and Google provide more context and specificity.

Apple is reportedly partnering with LG to launch a smart lock, thermostat, and doorbell, integrating with Apple's upcoming smart home hub, and competing with Amazon's smart home offerings.

ASOS customers received a strange notification claiming the company's Snowflake instance was compromised. The message, allegedly from a group called 'Xuanye Group,' threatened to leak data if ASOS didn't engage with them. ASOS confirmed unauthorized access to third-party platforms and is investigating.

OpenAI announced plans to introduce new visual ad formats in ChatGPT, including image-based ads. The company aims to provide more opportunities for advertisers to reach users, while also generating revenue. The ads will initially appear alongside image-generated content and may expand to other areas of the platform. Users who subscribe to ChatGPT Plus, Pro, or Enterprise plans will be exempt from seeing these ads.

A browser-based version of Grand Theft Auto V has been made available, showcasing how AI can be used to illegally port games to new platforms. The game's source code was leaked in 2023, and it appears that AI was used to build the game for WebAssembly and WebGPU, allowing it to run in a web browser. The site hosting the game was quickly taken down by Rockstar, but it highlights the potential for AI to be used in game development and porting.