24-hour AI news: regulatory refinement and corporate mergers and acquisitions go hand in hand, and the agent track is heating up

24-hour AI news at a glance: Policies and standards are "double landed", domestic agents and overseas mergers and acquisitions have accelerated

at the same frequency In the past 24 hours, deduplication checks have been completed: there has been no duplication of the same topic in the past 72 hours (the latest is August 31) after searching for designated channels, so only new progress is included in this issue. Maintain balanced coverage of domestic and foreign news, focusing on regulation and standards, corporate mergers and acquisitions and financial reports, and new products and agents.

1. OpenAI acquired Statsig for about $1.1 billion, and the application line formed an "Applications CTO" to coordinate the landing

1. OpenAI confirmed the acquisition of Statsig, a product experimental platform, for a transaction consideration of about $1.1 billion (all shares), and appointed founder Vijaye Raji as "Applications" Line CTO.
2. The new architecture focuses on application-side engineering and iteration such as ChatGPT, and strengthens A/B experiments, grayscale releases, and compliance pipelines.
3. Significance: Strengthen productization and ecological capabilities, paving the way for the release of multi-application matrix and enterprise-level capabilities in the next stage (overseas).

2. The two departments issued an action plan for the stable growth of the electronic information manufacturing industry from 2025 to 2026, proposing a "chip-large model adaptation test"

1. On September 4, the Ministry of Industry and Information Technology and the State Administration for Market Regulation issued an action plan, clarifying that "carry out adaptability testing of artificial intelligence chips and large models".
2. It is proposed to promote the innovative application of AI terminals, key devices and chip modules, and firmly promote the "domestic products and national use".
3. Significance: Establish a software and hardware collaborative verification path to help the domestic computing power ecology and large model industrialization (domestic).

3. Apple promotes Siri's "AI search/answer engine" and cooperates with Google Gemini evaluation (according to sources)

1. According to a number of media reports, Apple's internal AI search function, codenamed "World Knowledge Answers", is advancing for Siri, with a time target of spring 2026.
2. It is reported that Apple is evaluating access to Google Gemini to enhance general knowledge and multimodal responses, and may be expanded to Safari and Spotlight in the future.
3. Significance: The integration of voice assistant and AI search is accelerating, and the "model-terminal-entrance" linkage upgrade between giants (overseas, according to the news).

4. National Standards Committee: It is planned to formulate and revise more than 4,000 national standards around artificial intelligence and other fields

1. On September 4, it was reported that more than 600 national standards related to the transformation and upgrading of the manufacturing industry have been released during the year.
2. In the next step, focus on formulating and revising more than 4,000 national standards in the fields of artificial intelligence, Internet of Things, energy efficiency, etc., and strengthen implementation monitoring and policy coordination.
3. Significance: Standards first consolidate the "foundation" of industrial governance and application (domestic).

5. HPE financial report: Q3 revenue of $9.1 billion, AI system revenue of $1.6 billion, backlog of $3.7 billion

1. HPE announced financial report: Q3 total revenue of $9.1 billion (+19% year-on-year), non-GAAP earnings per share of $0.44.
2. AI Systems' single-quarter revenue was $1.6 billion (a significant increase from $1 billion in Q2), and the backlog of AI orders was about $3.7 billion; The annual AI system revenue target is about $4.6 billion.
3. The company reminded that due to the pace of large order delivery, Q4 AI-related revenue may fall by more than 30% month-on-month (overseas).

6. DeepSeek is said to be sprinting to release an "AI agent" model before the end of the year

1. On September 4, a number of media quoted people familiar with the matter as saying that DeepSeek plans to release an agent model with multi-step execution and self-improvement capabilities by the end of 2025.
2. The model emphasizes the "least instruction-multi-step" agent capability and long-term task execution, and positions the entrance to the next generation of human-computer interaction.
3. Positioning: It is "news/to be confirmed", but it shows the acceleration rhythm of domestic manufacturers in the Agent track (domestic, sources).

7. The European Commission launched the "Transparent AI System" guidelines and industry self-discipline guidelines to solicit opinions

1. On September 4, the European Commission launched a consultation to formulate guidelines for generative AI content detection and labeling, and promote relevant industry self-discipline guidelines.
2. This initiative is linked to the General Purpose AI (GPAI) rules in the AI Act (which came into effect on August 2) to clarify labeling, copyright and security requirements.
3. Significance: Global content label governance has entered the stage of "rule refinement + industry self-discipline" (overseas, supervision).

8. Huawei released the tri-fold flagship Mate XTs, and the system-level agent "Xiaoyi" was fully upgraded

1. September 4 press conference: Mate XTs debuted with Extraordinary Master, equipped with HarmonyOS 5.1, with a starting price of 17,999 yuan (RMB).
2. Xiaoyi knowledge base and "in-depth problem-solving agent" and other capabilities are new, focusing on multi-modal and long-link agent experience in business efficiency and education scenarios.
3. Significance: The combination of high-end hardware + system-level agent promotes the "resident" landing of AI in terminals (domestic).

9. C3.ai fell more than 10% in pre-market trading, and performance and guidance were under pressure and CEO replacement

occurred 1. C3.ai announced fiscal quarter revenue of about $70.3 million, issuing weaker follow-up guidance than the market, and the stock price fell more than 13% before the market.

1. Management changes: Stephen Ehikian takes over as CEO; Founder Tom Siebel stepped down due to health problems (impaired vision).
2. Significance: The challenge of "growth-monetization-delivery rhythm" of the enterprise-level AI software track still exists (overseas).

10. Beijing Lier joins hands with SenseTime and Xiwang to explore vertical models and agent applications in industrial manufacturing

1. Announced on the evening of September 4: The three parties intend to carry out strategic cooperation in the direction of AI computing power cooperation, industrial manufacturing and decision-making AI vertical model development, and related agent promotion.
2. The company reminds that there is still uncertainty about the downstream demand for this new business.
3. Significance: The combination of the material industry and AI models/agents further penetrates into in-depth scenarios (domestic, application landing).

Frequently Asked Questions (Q&A)

Q: What has changed in OpenAI's acquisition of Statsig?

A: The acquisition amount is about US$1.1 billion, and the founder is the CTO of the Applications line, which means that the experimentation, release and compliance pipeline of applications such as ChatGPT will be significantly accelerated, and enterprise-level product capabilities and ecological construction will be strengthened simultaneously.

Q: How far has the domestic "AI content labeling" advanced?

A: The marking measures and supporting mandatory national standards have come into effect on September 1; On September 4, the authoritative interpretation emphasized the "dual-track system" of explicit/implicit identification, and platforms such as WeChat and Douyin have issued implementation rules and entered the implementation stage.

Q: Is Apple's partnership with Google Gemini real, and when will it be available?

A: It belongs to the "news/foreign media report" stage: Apple's evaluation uses Gemini to enhance Siri's "answer engine", and the target time points to the spring of 2026.

Q: What does HPE's data mean for the AI infrastructure cycle?

A: Q3 AI system revenue is $1.6 billion, backlog 3$700 million shows strong demand, but Q4 may fall by 30%+ due to the rhythm of large orders, suggesting that "batch/project" fluctuations are still there - long-term orders and backlogs are more indicative.

Q: If DeepSeek releases "agent" at the end of the year, how will it be different from the previous R1 inference model?

A: The focus of R&D has shifted from "answer quality/reasoning" to "multi-step execution/long-term task/environmental interaction", with more emphasis on completing complex task chains with minimal instructions and positioning the next generation of human-computer collaboration interfaces (it is still reported that there is uncertainty at the time point).