AI Industry Development & Policy Dynamics

On September 1, CrowdStrike and Nvidia launched SafeMind, a family of AI security models and harnesses built for CrowdStrike's Falcon platform. The release includes Red Tempest for offensive testing and Blue Solano for defensive operations, trained on Falcon telemetry, threat intelligence, and

According to 动察 Beating, Anthropic has added context locks to Claude's encrypted reasoning in Fable 5.1. Encrypted thoughts generated via the API must now be returned alongside the original system prompt, tools, and message history. If any preceding content is modified, the API returns an error or d

Since August 30, Anthropic has notified some Claude users that infostealing malware was used to steal Claude login session cookies from their computers, bypass 2FA to access their accounts, and consume their usage. Anthropic clarified that the Claude platform itself was not breached; Claude accounts were merely used as a stepping stone for monetization. Anthropic Notification: Abnormal Sessions Logged Out In its notification email, Anthropic explained that it took the following measures after de

BitMEX co-founder Arthur Hayes announced on X on September 1 that Technocore.chat, developed by Flop Labs, had officially launched. It is a chat and notes server for AI agents that requires no login, client, or JavaScript, allowing users to post messages, read messages, and write notes using a single HTTP GET request. Technocore.chat’s Technical Design According to technical documentation published by Flop Labs on GitHub, most AI agents operate in sandboxed environments where the only authorized

NVIDIA and MediaTek’s jointly developed RTX Spark AI PC has sold out its first batch of inventory at initial partner manufacturers ASUS and MSI during the presale period, with both manufacturers actively requesting additional allocations from NVIDIA. RTX Spark is built using TSMC’s 3nm process and combines an NVIDIA Blackwell RTX GPU with a 20-core Grace CPU. NVIDIA RTX Spark Specifications RTX Spark is built using TSMC’s 3nm process and integrates an NVIDIA Blackwell RTX GPU with a 20-core Grac

According to Alibaba, Qwen3.8-Max received an upgrade on September 2, introducing Qwen3.8-Max-0902 with improved coding performance. The model maintains 2.4 trillion parameters and 1 million token context window. Across 8 programming benchmarks, all scores increased: TerminalBench 3.0 rose

Crypto treasury company BitMine (BMNR) announced its latest asset allocation data on August 31, with total assets reaching $15.6 billion and total ETH holdings reaching 5,901,112 ETH. BitMine stock rose 43% over the past month. Chairman Tom Lee said that South Korean investors have resumed buying cryptocurrency and selling AI stocks, while the CLARITY Act is a recent catalyst. BitMine Asset Breakdown: Total Assets of $15.6 Billion As of August 30, 2026, BitMine’s primary assets were as follows:

According to CNBC, SB Energy, backed by SoftBank, OpenAI, and Nvidia, filed for a US initial public offering on September 1, disclosing a net loss of about $3.2 billion on revenue of about $139 million in the first half of 2026. The company said none of its data centers are operational yet and

Cybersecurity company Palo Alto Networks (NASDAQ: PANW) released its FY2026 Q4 results on Sept. 1 and simultaneously announced the acquisition of AI-native platform Console. Q4 revenue was $3.41 billion, up 34% year over year and above analysts’ expectation of $3.35 billion; adjusted EPS was $1.02, above the expected $0.98. FY2026 Q4 Results Palo Alto Networks’ key financial results for FY2026 Q4 were as follows: revenue of $3.41 billion, up 34% year over year and exceeding analysts’ expectation

According to LS Securities, Hyundai Mobis secured full orders for actuators for Boston Dynamics' Atlas humanoid robot on September 2, with 31 units per robot. Actuators are core drive components that control the robot's joints and represent the largest cost component in humanoid manufacturing. The a

According to ChainCatcher, Chinese AI startup Zhipu opened an official flagship store on Tmall today (September 2), selling Coding Plan subscription packages. Available tiers include Lite, Pro, and Max for individual users, with monthly, quarterly, and annual subscription options. The company

Alibaba updated its flagship Qwen3.8-Max model on September 2, with the latest version achieving first place globally in frontend coding ability. On the CodeArena benchmark, a leading third-party evaluation platform for web development, Qwen3.8-Max improved 22 points to 1691, surpassing Claude

According to Alibaba Cloud, the company backed Malaysia's AI untuk Rakyat program on September 2, providing its MuleRun workflow automation platform and WonderClip AI video creation tool through the initiative. The program, implemented by MyDIGITAL Corporation under Malaysia's Ministry of Digital, targets 100,000 Malaysians aged 18 to 30 and aims to expand AI access and learning opportunities for young people. Participants who complete and pass required learning modules can redeem three months o

On Tuesday (Sept. 1), Anthropic launched Claude 5.1, cutting cache read fees by 75% to $0.25 per million tokens while maintaining input and output token prices. The company said typical workloads would see costs decline by about 25%, with complex agent-based tasks saving up to 45%. Claude 5.1

Tencent released and open-sourced “Hunyuan Hy4 Preview” on Aug. 31. It uses a mixture-of-experts (MoE) architecture with 770 billion total parameters and a context window exceeding 1 million tokens. In a blind test involving 203 engineering tasks and 163 Tencent experts, Hy4 Preview averaged 2.99/4.00, slightly outperforming GLM-5.3 and Kimi K3. Core Specifications of Hy4 Preview: 770-Billion-Parameter MoE Architecture According to Tencent’s release materials, Hy4 Preview uses a mixture-of-exper

According to ChainCatcher, Google announced on September 1 at SEMICON Taiwan that high-performance memory now accounts for 75% of AI server hardware costs. To address memory bottlenecks, Google introduced the TPU 8i for low-latency inference with 288 GB of high-bandwidth memory and 384 MiB of

Gate News reported that, as of press time on September 2, UAI (AImerica) was trading at $0.55, up 52.67% over the past 24 hours, according to Gate market data. The token reached a high of $0.61 and a low of $0.38, while 24-hour trading volume reached $1.9408 million. Its current market cap is approximately $131 million. UnifAI is purpose-built artificial intelligence infrastructure for intelligent finance, supporting next-generation autonomous AI agents. By providing a highly available distribut

Anthropic signed a $35 billion cloud computing agreement with Lambda, a cloud computing provider backed by Anthropic and NVIDIA, with some of the capacity located at the Beacon Point campus being built by Hut 8 (HUT) in Nueces County, Texas. HUT shares initially rose after the news emerged before retreating, closing at $77.57 on Sept. 1. Anthropic and Lambda’s $35 Billion Agreement Anthropic signed a $35 billion agreement with NVIDIA-backed cloud computing provider Lambda to build data center co

According to Foresight News, Figure (Nasdaq: FIGR) has completed its acquisition of AI-powered real estate lending platform Kiavi for $717 million. Figure acquired Kiavi's technology, operating platform, and certain other assets. The platform and technology will be integrated into Figure's network o

According to the China Open-Source LLM Tracker published by Robonomics author FD on September 1, the core large language model revenue of China's open-source sector reached 60-80 billion dollars, with total ARR standing at 100-150 billion dollars after excluding ByteDance and Seedance. The sector's