24-Hour AI News: California AI Law Enacted and Claude 4.5 Launched, Domestic Policies and Open Source Promoted

In the past 24 hours, China released and implemented "AI+" policy tools, leading to the implementation of these tools in various industries. Zhipu open-sourced GLM-4.6. Overseas, California signed the AI Security Disclosure Act, Anthropic released Claude Sonnet 4.5, and OpenAI's revenue and product development trends coincided. This channel has excluded stories already reported on the same topic within the past 72 hours, and only includes content with new developments.

1. California signed the AI Safety Disclosure Act (SB 53), requiring leading companies to disclose "catastrophic risk" mitigation plans

1. It was signed and took effect on September 29, covering companies with revenue ≥ US$500 million, including OpenAI, Google, Meta, Nvidia, Anthropic, etc.
2. High risks such as loss of control and dual use must be assessed and disclosed to the public. The maximum fine for a single violation is US$1 million.
3. It marks the continued advancement of AI governance at the state level in the United States, and may promote the process of unified rules at the federal level.

Anthropic Releases Claude Sonnet 4.5: Significant Improvements in Long-Term Autonomous and Enterprise-Level Coding Capabilities

1. Officials claim that the system has been comprehensively upgraded in terms of coding, reasoning, and "computer-using" capabilities, and can run tasks autonomously for 30 hours continuously.
2. Simultaneously enter the Google Vertex AI, AWS Bedrock and Microsoft Copilot ecosystems, targeting enterprise-level and complex intelligent scenarios.
3. Third-party evaluations and developer practices show significant improvements over previous generations in tasks such as SWE-bench, with enhanced stability for long-term tasks.

3. OpenAI's revenue in the first half of the year was approximately US$4.3 billion, an increase of approximately 16% over the same period last year.

1. The revenue structure is driven by the enterprise version and API, indicating an increase in B2B penetration.
2. According to media reports, its short video generation and distribution application has entered the internal testing preparation stage and may be bound to Sora 2 capabilities.
3. Copyright and content compliance continue to receive attention, and video generation will reportedly adopt an "opt-out" strategy for rights holders.

4. National Development and Reform Commission: The total scale of new policy-based financial instruments is 500 billion yuan, emphasizing support for the implementation of "artificial intelligence +"

1. The press conference on September 29th made it clear that the entire 500 billion yuan would be used to supplement project capital and speed up investment in major projects.
2. Emphasize the promotion of the popularization of the new generation of smart terminals and smart bodies, and set a goal of exceeding 70% (caliber) in penetration rate by 2027.
3. Release the synergistic signal of "computing power-scenario-capital" to promote the large-scale deployment of AI applications.

5. Zhipu Releases and Open-Sources GLM-4.6: Enhanced Long Context and Agentic Coding Capabilities

1. Announcement of open source and domestic chip adaptation: FP8+INT4 hybrid quantitative inference was put into production in Cambricon, and vLLM adaptation was completed on Moore's Threads.
2. The context length is increased to approximately 200KB, and code, reasoning, and agent applications are multi-dimensionally enhanced.
3. It has been interpreted by many media as "a new step in the open source ecology of domestic models" and has exemplary significance for the "model-core collaboration" of the industrial chain.

6. Google's "AI Mode" adds visual results and expands the generative search experience

1. AI-driven search questions and answers have expanded from text to image-level results, covering shopping and inspiration scenarios.
2. Collaborate with the Gemini ecosystem to complement visual narrative generation and enhance interactive visualization.
3. Compare competitors’ multimodal search and content generation to further accelerate the trend of “search-creation” integration.

7. Shenzhen Qianhai issues AI support application guide, opening a new round of application acceptance window

1. The application period is from September 30 to November 6, focusing on AI companies that actually operate in the district.
2. Clarify requirements for large-scale model and algorithm filing, computing power, and scenario applications.
3. Combined with the “artificial intelligence +” policy direction, it is expected to form a Bay Area-level industrial taking-over effect.

8. Beijing's "Analog World" ecosystem adds 8 new companies, creating a "1km AI industry core ecosystem"

1. Form an integrated ecosystem of computing power, data, scenarios, investment and financing, and policy services.
2. The first batch of “more than 30” high-quality projects are “full park as soon as they open”, emphasizing full life cycle incubation and industrial collaboration.
3. Assist in the development of regional AI innovation clusters and the cultivation of “new quality productivity”.

9. Honor opened the world's first "Alpha Flagship Store" in Shenzhen, strengthening the offline implementation of the AI terminal ecosystem

1. Positioned as the "AI Terminal Smart Life Innovation Workshop", it is aimed at co-creation between creators and partners.
2. Demonstrate the implementation of AI in homes and mobile devices through scenario-based solutions.
3. Terminal brands accelerate the popularization of AI terminals through "offline experience + ecological interoperability".

10. Inspur Information showcased two AI server systems at AICC, focusing on low latency and low-cost inference.

1. Released the MetaBrain SD200 super node and HC1000 ultra-scalable AI server.
2. Propose the "10 milliseconds, 1 dollar" level of reasoning goal (caliber), emphasizing engineering capabilities for the era of intelligent agents.
3. Cooperate with ecological partners to promote the collaboration of "full-stack intelligent computing + industry applications".

Frequently Asked Questions (Q&A)

Q: Which event in these 24 hours has the greatest impact on the direction of the industry?

A: The signing of California SB 53 and the release of Claude Sonnet 4.5 by Anthropic are both key. The former sets a model for "state-level mandatory risk disclosure," while the latter represents significant progress in long-term autonomous and enterprise-level coding, which will directly impact enterprise deployment rhythms and compliance expectations.

Q: What are the signals of “real money” landing in China?

A: The National Development and Reform Commission has clearly stated that 500 billion yuan of policy-based financial instruments will be used to supplement project capital, and emphasized the goal of popularizing "artificial intelligence +" and intelligent bodies; Shenzhen Qianhai has opened a new round of AI support application window, with funds and scenarios working in both directions.

Q: Why is the new progress of DeepSeek not highlighted in this issue?

A: This channel has already covered "DeepSeek Iteration and Price Reduction" on September 29. To avoid repetition, this issue will only mention it where comparisons or ecological context are involved, and will no longer be listed separately unless new quantitative milestones appear.

Q: With so many models released, how should companies choose?

A: Prioritize three factors: 1. Compliance and data sovereignty (whether there is a clear commitment to security and compliance); 2. Task fit (whether long context, agentic coding, tool usage capabilities, etc. meet the business needs); 3. Ecosystem availability (whether it is integrated with cloud vendors and office suites to facilitate rapid implementation).