AI
AI News: China's $295B Buildout, Claude Finds Zcash Bug, and More
China commits $295B to domestic AI chips, Claude uncovers a critical Zcash flaw, and Google's NotebookLM gains code execution.
China Plans $295 Billion AI Infrastructure Investment, Prioritizing Domestic Chips
Beijing is preparing a five-year, 2 trillion yuan ($295 billion) investment to build a nationwide data center network, with at least 80% of technology sourced from domestic suppliers like Huawei, effectively locking out Nvidia and AMD. The plan, part of the ‘Six Networks’ program, aims to connect scattered data centers into a unified system by 2028. Meanwhile, Taiwan is considering criminalizing the export of AI chips to China for the first time.
Google’s NotebookLM Upgraded with Gemini 3.5, Gains Code Execution and Advanced Data Analysis
NotebookLM now runs on the Gemini 3.5 model, enabling secure cloud-based code execution and advanced data analysis. Users can clean datasets, automate workflows, and generate outputs in formats like PDF, Excel, and PowerPoint directly within the platform. The update introduces guided workflows for organizing source libraries and integrating Google Search materials, reducing the need for multiple tools.
Why NotebookLM’s Gemini 3.5 Upgrade is a Major Shift for Data Analysis and Research →
Stealth Startup ‘A’ Raises $37 Million for AI-Powered Autonomous Cyberattacks
The New York-based startup ‘A’ emerged from stealth with $37 million in funding from Lightspeed Venture Partners, Cyberstarts, and executives from Wiz and Cyera. Its AI agent continuously breaks into customers’ systems to find and fix vulnerabilities before real attackers can exploit them. In a proof of concept, it uncovered 1.2 million exposed sensitive records that had gone undetected for years.
Why Lightspeed and Wiz’s Assaf Rappaport bet $37 million on an AI-powered cyberattacker →
OpenClaw Vulnerability Exposes Risks of Unmanaged AI Agents in Enterprise
Oasis Security discovered a critical vulnerability in the popular AI agent OpenClaw, allowing any website to silently take control of the local agent via its WebSocket gateway without user action. The flaw highlights the governance gap as 74% of companies plan to deploy agentic AI within two years, yet only 21% have mature oversight. OpenClaw’s broad permissions and local operation make it a prime target for exploitation.
What the OpenClaw vulnerability reveals about the future of agentic AI security →
Claude Opus 4.8 Finds Critical Zcash Bug, Sending Price Down 40%
Within 24 hours of public release, Anthropic’s Claude Opus 4.8 helped a security researcher discover a four-year-old vulnerability in Zcash that could allow unlimited counterfeit coin minting. The bug disclosure caused Zcash to drop 40% overnight. The event raises concerns for Ethereum and DeFi, as AI models can now identify flaws that traditional audits missed, potentially increasing risk for smart contract-based systems.
📡 From social media
John Hammond Launches OnlyLANs, a Free Prompt Injection Wargame
John Hammond introduced OnlyLANs, a free prompt injection wargame designed to help users troubleshoot local networks with an AI assistant while challenging them to leak secrets. The game, developed by Andrew Bellini and Eva Benn for ContinuumCon 2026, offers prizes including ticket discount codes and raffle entries for completing Level 1. OnlyLANs is part of the Just Hacking Training community projects and follows the release of a Windows Malware Development course.