美国AI标准机构：Kimi K3网络安全能力落后于美国前沿模型，安全防护仍允许漏洞利用开发
Related
Hubei Provincial Industrial Software Manufacturing Innovation Center Established
Mars Finance reported on June 11 that the 4th Software Innovation and Development Conference was held in Wuhan. It was learned at the event that the Hubei Provincial Industrial Software Manufacturing Innovation Center was simultaneously inaugurated, integrating resources from industry, academia, research, application, and finance across the province to create a public innovation platform encompassing technological breakthroughs, standards development,成果转化 (technology transfer), and talent cultivation. A relevant official from the Hubei Provincial Department of Economy and Information Technology pointed out that Hubei Province will next study and introduce a series of action plans, including "Artificial Intelligence + Software," to strengthen its two distinctive clusters: independent and secure computing and industrial software; focusing on independent controllability, deepening the "first-version" software policy, and exploring the application of "industrial software vouchers." (Securities Times)
Shanghai Stock Exchange: Companies producing large-scale AI models are eligible for the fifth set of listing standards on the Science and Technology Innovation Board.
According to Mars Finance, the Shanghai Stock Exchange (SSE) announced that, in order to implement the "Opinions of the China Securities Regulatory Commission on Setting Up a Science and Technology Innovation Growth Layer on the Science and Technology Innovation Board to Enhance the Inclusiveness and Adaptability of the System," further standardize the application of the fifth set of listing standards for technology-based enterprises on the Science and Technology Innovation Board, and support high-quality artificial intelligence large-scale model enterprises that have not yet formed a certain revenue scale to issue and list on the Science and Technology Innovation Board, the SSE has formulated the "Shanghai Stock Exchange Issuance and Listing Review Rules Application Guidelines No. 10—Application of the Fifth Set of Listing Standards for Artificial Intelligence Large-Scale Model Enterprises on the Science and Technology Innovation Board," which is hereby issued and will take effect from the date of issuance. (Cailian Press)
The National-Local Jointly-Built Humanoid Robot Innovation Center and the Robot Lifecycle Demonstration Application Zone were launched.
According to Mars Finance, the 2026 Zhangjiang Embossed Intelligent Supply Chain Conference was held in Shanghai on June 16. At the opening ceremony, the National-Local Jointly Constructed Humanoid Robot Innovation Center's Robot Lifecycle Demonstration Application Zone was launched. Jiang Lei, Chief Scientist of the National-Local Jointly Constructed Humanoid Robot Innovation Center, told a reporter from the Securities Times, "The industry generally believes that embodied intelligent robots are a field that requires 10-20 years of in-depth development. We hope to compress this development process to 5-8 years." He said that the demonstration application zone will draw on the experience of full lifecycle management in high-tech industries, establishing a full lifecycle management system in the early stages of the robot industry's development, covering the entire cycle of robots from core component production and embodied intelligent algorithm implantation to finished product delivery to customers and routine maintenance. (Securities Times)
The White House plans to release new AI regulatory standards next week, retracting Trump's previous promise of "light regulation."
According to Beating's monitoring, the Trump administration's "light regulation" approach to AI is facing serious challenges. The Financial Times reports that the White House will release a set of voluntary standards for the release of cutting-edge AI models as early as next week. These standards aim to set benchmarks for models with top-tier cybersecurity capabilities and establish a release timeline to regulate future release processes. The Center for AI Standards and Innovation and the National Security Agency (NSA) will play key roles in standard setting and oversight. Last month, the Trump administration's direct intervention in two major AI labs triggered industry chaos and anger: Anthropic's latest model was subject to export controls on June 12 due to concerns about misuse and cybersecurity, and was only unbanned this Tuesday; OpenAI was required to release GPT-5.6 only to a government-selected population, with a wider release expected as early as next week; Google is also communicating with the government before releasing a new generation of coding models with stronger cybersecurity capabilities. Although Trump promised to ease regulations so that American companies could outperform their Chinese counterparts, the potential risks of new models exploiting security vulnerabilities and disrupting key industries are forcing the White House to accelerate the tightening of security measures.
Kimi's B-end head: The AI industry has a bubble but solid fundamentals; API price increases stem from computing power shortages.
According to ChainCatcher, citing 21st Century Business Herald, Huang Zhenxin, head of the B2B division at Moonshot AI's Kimi, stated at a recent communication meeting that while there is indeed a bubble in the AI industry, its fundamentals are very solid. Enterprises can now clearly calculate their return on investment (ROI), and AI has brought about a substantial qualitative change in productivity. Regarding the recent widespread price increases by model vendors, Huang Zhenxin pointed out that the core reason is the rising global computing power costs and the inability of chip production capacity to meet the explosive growth in token demand. He emphasized that evaluating the cost-effectiveness of a model should not only consider the unit price of inputs and outputs, but should focus on the cache hit rate. It is understood that Kimi's original cache hit rate has reached over 90%, significantly reducing actual computing costs. Furthermore, Huang Zhenxin revealed that Kimi will continue to challenge underlying architectural innovation to perpetuate the Scaling Law, and its Muon optimizer, which was first validated on a large scale, is now widely used by many mainstream large-scale models in the industry. Regarding the "last mile" of enterprise AI implementation, he believes that as the basic capabilities of models continue to improve, the technical paradigm of the application layer will also continue to simplify.
Kimi is seeking to raise up to $2 billion at a valuation of $30 billion.
Odaily Odaily reports that AI startup Moonshot AI is seeking to raise up to $2 billion in a new funding round, which would value the company at $30 billion. Moonshot AI is currently in early-stage talks with potential investors and plans to raise more than $1 billion. Meanwhile, the company is nearing completion of a $2 billion funding round led by Meituan. As of April, Moonshot AI's annual recurring revenue had exceeded $200 million, driven by increased demand for its Kimi chatbot and large language models. Furthermore, the company is dismantling its offshore structure to pave the way for an initial public offering in Hong Kong and plans to establish a joint venture structure to allow foreign investment. The company recently launched Kimi Work, a general-purpose AI agent based on its K2.6 series models. (Bloomberg)