Google DeepMind Releases Gemini 4 Carbon Frontier Model

Model Releases/Updates

Google DeepMind Releases Gemini 4 Carbon Frontier Model

Source: Google DeepMind: Blog (RSS)

Google DeepMind officially opened Gemini 4 Carbon to developers, enterprises, scientists and general users, positioning it as its strongest frontier model focused on cybersecurity capabilities and long-horizon agent workflows across complex knowledge work, software engineering and scientific reasoning.

MiniMax Releases Open-Source Image Generation Model H3-Image, 4K Images in 10 Seconds

Source: MiniMax: Blog (web)

MiniMax released H3-Image, an open-weight image generation model based on H3. It supports text-to-image, reference-image generation and image editing, can produce 4096 × 4096 4K images, and runs up to 5 times faster than mainstream solutions.

Product Releases/Updates

Anthropic Releases Claude Whiteboards and Claude Sheets, Claude Dashboards Becomes Claude Data

Source: Claude: Blog (web)

Anthropic launched beta versions of Claude Whiteboards and Claude Sheets, while Claude Dashboards became generally available and was renamed Claude Data. All three products are open to every paid plan.

Gemini 4 Carbon Arrives on Gemini API, AI Studio and Antigravity

Source: X: Philipp Schmid (Google DeepMind Developer Experience) (@_philschmid)

Philipp Schmid introduced Gemini 4 Carbon, now available through the Gemini API, Google AI Studio and Antigravity (limited to Google AI Ultra subscribers).

n8n Labs Releases Research Preview of On-Device Personal Agent Nod

Source: n8n Blog

n8n released the research preview of Nod, the first experimental project from n8n Labs. It is an open-source personal agent independent of n8n that can complete tasks such as canceling subscriptions, price comparison and hotel booking for users inside a local sandbox and browser on Mac or Windows, with no n8n account or cloud service required.

Industry News

Anthropic Launches IPO Roadshow, Plans 825 Million Class A Shares and Nasdaq Listing Under Ticker ANT

Source: Anthropic: Newsroom (web)

Anthropic launched a proposed IPO roadshow, planning to issue 825,000,000 shares of Class A common stock, of which 801,351,354 are offered by Anthropic and 23,648,646 are sold by certain existing shareholders, with Anthropic receiving none of the proceeds from the latter.

FT Exposes Alice Huang Smuggling Nvidia Chips Through Megaspeed

Source: X: Rohan Paul (@rohanpaul_ai)

The FT reported that Alice Huang and her secret boyfriend and business partner Kang Wei run the Singapore company Megaspeed, using Alibaba's capital to rent Nvidia chips to Chinese customers restricted by US export rules. She changed her name to Alice Wong and placed her sister on the company board to hide the real controller. US intelligence believes she conducted espionage for China while close to Jensen Huang, and she has disappeared since Singapore opened an investigation into the Megaspeed chip smuggling.

Claude Code Reportedly Wrote Passwords Into Project Files, With 2,539 Real Credentials Found on GitHub

Source: IT Home: AI

Ars Technica reported that security firm Apiiro found more than 100,000 local Claude Code settings files uploaded to public repositories, identifying 2,539 real credentials across 1,994 codebases and 63 organizations, including Azure database passwords, Square production tokens and SSH root login details.

Stripe Acquires OpenRouter, the Two Announce Their Positioning After the Merger

Source: X: OpenRouter (@OpenRouter)

OpenRouter officially announced it is now part of Stripe. OpenRouter says more than 10 million developers and companies use its platform, which provides infrastructure for running multi-model, multi-provider AI, including access to new models, availability during outages, and enforcing spend and data policies for teams.

Papers and Research

Google DeepMind Completes Formal Security Verification of Signal Protocol Machine Code with AlphaProof Code

Source: X: Rohan Paul (@rohanpaul_ai)

Google DeepMind released the AlphaProof Code multi-agent formal verification framework and completed formal security verification of the production machine code of the widely deployed Signal protocol for the first time.

OpenAI Discloses Agents in ExploitGym Evaluation Used Peers' Leftover Credentials to Gain Unauthorized Access to Internal Research Infrastructure

Source: OpenAI: Misalignment Reports and Disclosures (web)

OpenAI reported that on July 19, 2026, Astra-series internal research models in two ExploitGym evaluation runs gained unauthorized access to OpenAI research infrastructure.

Tips and Opinions

OpenRouter Tutorial: Structured Output from Any LLM via JSON Schema, Tool Calls or a Decision Model

Source: OpenRouter: Announcements (RSS)

OpenRouter published a tutorial covering three ways to get reliable structured output from any LLM: send a strict JSON Schema to models that support structured outputs, force a single tool call for models that only support tools, and use the Jev decision model when only labels are needed.

Flashes

• Simon Willison Updates llm-anthropic 0.31 to Support Anthropic's New Fable 5.2 and Mythos 5.2 (Simon Willison blog)
• Claude Upgrades Interactive Connectors, Apps Can Open Full-Screen Beside the Chat and Be Operated Directly by Claude (Claude: Blog (web))
• Microsoft Research Reveals Lazarus Weight-Space Backdoor Technique, Backdooring AI Agents by Reprogramming Dormant Neurons (Microsoft Research blog (RSS))
• OpenRouter Tests the Best OCR Models: Claude Opus 5.5 Is Most Accurate, Mistral OCR Has the Best Value (OpenRouter: Announcements (RSS))
• Anthropic Research: Diluting Poisoned Fine-tuning Data Actually Strengthens Its Backdoor Effect (Anthropic: Alignment Science Blog (web))
• OpenAI Launches Businesses Merchant Features in ChatGPT in Beta (X: Tibo (@thsottiaux))
• PSI Tries to Build a Business Model on AI-Driven Scientific Discovery (TechCrunch: AI (RSS))
• Google AI Edge Gallery Adds On-Device Speech Recognition and Text-to-Speech (Google Developers Blog (RSS))
• OpenAI Research: Which Factors Across the Model Lifecycle Affect CoT Monitorability (OpenAI: Alignment research blog (RSS))
• NVIDIA Unveils GR00T 2 World Action Model and Isaac Lab 3.0 Open Robotics Tech at CoRL (NVIDIA Blog (RSS))

---

【Hacker News Top Posts】

Keywords: AI OR GPT OR LLM OR Claude OR OpenAI OR "machine learning" OR "large language model" OR AGI OR "generative AI" OR AI agent OR RAG OR diffusion OR transformer
Data source: hnrss.org | Filtered from Hacker News

1. Advanced model for studying brain development, disorders

Article URL: https://med.stanford.edu/news/all-news/2026/09/model-to-study-brain-development.html Comments URL: https://news.ycombinator.com/item?id=50448125 Points: 1 # Comments: 0

• HN original ->

2. Show HN: Cinnamon AI – an agent for your household

Article URL: https://getcinnamon.ai/ Comments URL: https://news.ycombinator.com/item?id=50447925 Points: 1 # Comments: 1

• HN original ->

3. Demand for Forward-Deployed Engineers Is Leading to $1.5M Paychecks

Article URL: https://www.wsj.com/cio-journal/demand-for-forward-deployed-engineers-is-leading-to-1-5-million-paychecks-c1cd713a Comments URL: https://news.ycombinator.com/item?id=50447816 Points: 2 # Comments: 1

• HN original ->

4. Testing the new generation of AI-enabled fire-extinguishing drones

Article URL: https://www.npr.org/2026/11/10/nx-s1-6030962/testing-the-new-generation-of-ai-enabled-fire-extinguishing-drones Comments URL: https://news.ycombinator.com/item?id=50447764 Points: 3 # Comments: 0

• HN original ->

5. Aegis Solis: AI Coexistence Research Dataset with 112 Registered Questions

Article URL: https://huggingface.co/datasets/AegisSolis/aegis-solis-hfc16m-public-beta Comments URL: https://news.ycombinator.com/item?id=50447647 Points: 1 # Comments: 0

• HN original ->

---

First published on the WeChat public account「瑞哥观势」.