In the past 24 hours (January 9 to January 10, 2026, Pacific time), domestic AI policy and compliance progress have been released intensively, and the industry side has moved from "usable" to "scalable"; Overseas, the trends of end-side AI and physical AI were focused on displaying at the closing node of CES, and discussions around advanced computing power exports, AI proxy evaluation and data security heated up.
1. Generative AI filing has accelerated, with 446 new services added throughout the year
The latest data shows that 446 new generative artificial intelligence services will be registered in 2025, and the cumulative number of services that have completed the filing will further increase. The clarity of the compliance threshold means that the pace of product launch and commercialization is more predictable, but the investment in content security and data governance will also become the "standard cost".
2. The special action of "artificial intelligence + manufacturing" has been introduced, and the quantitative goals by 2027 have been clarified
Multiple departments jointly issued a special action document, proposing the goal of launching 1,000 high-level industrial agents, creating 100 high-quality data sets in industrial fields, and promoting 500 typical application scenarios by 2027. The policy signal points to "application traction + data and computing power base", and industrial scenarios will become one of the main battlefields for the large-scale implementation of AI.
3. Cybersecurity supervision emphasizes "supporting AI development + strengthening security responsibilities"
Further clarify the requirements for network security legal responsibilities and key equipment security certification, and at the same time propose to support the research and development of key AI technologies and strengthen risk monitoring and assessment at the institutional level. For enterprises, security assessment, compliance traces, and supply chain security will be more deeply embedded in the entire life cycle of AI products.
4. Zhipu announced the open source of AutoGLM to promote the agent ecology of "operating mobile phones"
Following the release of the agent model that can operate the mobile phone, Zhipu further promoted the open source action. Open source is expected to drive faster application innovation and ecological collaboration, and at the same time, it will also make "agent controllability, permission boundaries and misoperation" an engineering problem that developers must face.
5. MiniMax and Zhipu have been listed on the Hong Kong Stock Exchange, and the capital market has priced the large model track
The two large model companies have completed their listings and attracted market attention, and the relevant transaction performance and market value changes are regarded as the vane of the industry's "from R&D competition to commercial realization". For startups and industrial chains, the financing logic may shift from "telling model stories" to "telling customers and cash flow".
6. CES 2026 ends: AI shifts from "model capabilities" to "product capabilities"
The exhibition observation shows that the focus of AI discussions is more on "what can be done, how to embed the system, and how to deploy at scale". Device-side AI and physical AI have become high-frequency keywords, which means that the importance of computing power, sensors, system integration and software engineering capabilities has increased significantly.
7. CES focuses on the release of AI-related chips and tools, and the hardware competition has risen to another level
A number of manufacturers launched new products and platform updates for AI workloads at the CES node, covering the PC, edge and consumer electronics ecosystems. For application developers, inference costs and device-side experience will improve faster, but the pressure of adaptation fragmentation and cross-platform optimization will also increase simultaneously.
8. Housework robots are "more like sellable products", and the verification of family scenarios is accelerated
At the exhibition, there were more robot displays for housework and escorts, and the ability moved from demonstration to a more complete task chain. Commercialization prospects are closer, but security standards, liability definitions, and family privacy will determine how quickly they will truly become popular.
9. The United States has strengthened legislation and law enforcement to restrict the sale of advanced AI chips to China
Discussions on the review and restrictions on the export of advanced computing power in the US political circles have heated up, and relevant bills and regulatory resource allocation have been put on the agenda. For global supply chains, uncertainty is likely to continue to rise, and companies need to prepare for multi-source supply and performance alternatives.
10. AI agent evaluation causes data and confidentiality disputes: how to use real work products compliantly
There are reports that to evaluate AI agent capabilities, some outsourced evaluation processes require the submission of real or simulated work samples, emphasizing the removal of sensitive information. As a result, the industry once again focuses on "evaluating data sources, desensitization effectiveness, and trade secret boundaries", and may promote more standardized data compliance processes in the future.
Frequently Asked Questions (Q&A)
Q: What are the most obvious industry keywords in the last 24 hours?
A: The key word is "implementation and governance in parallel", on the one hand, industrial policy and hardware promote large-scale application, and on the other hand, compliance, export control and data security write the risk boundaries more clearly.
Q: What is the difference between domestic and foreign focus?
A: China focuses more on "industrial integration and compliance mechanism", emphasizing manufacturing scenarios and filing management; Foreign countries are more focused on the game of "device-side and robot productization" and "advanced computing power and security risks".
Q: What are the most direct opportunities for developers and entrepreneurs?
A: Opportunities are focused on "deliverability" in industrial and end-side scenarios, including industry data engineering, system integration, task-based agents and inference optimization, and compliance-oriented content security and auditing tools.
Q: What is the biggest risk point in this period?
A: Risks mainly come from three types of uncertainties: computing power and supply chain constraints, permissions and security vulnerabilities caused by agents, and compliance and trade secret risks of evaluation and training data.
