24-hour AI news: Governance and security achievements are released in a centralized manner, and iOS 26 is released together with Samsung Forum

In the past 24 hours, China released the "Artificial Intelligence Security Governance Framework" 2.0, announced the results of AI-enabled network security tests, and launched an antitrust investigation against Nvidia; The Guangdong-Hong Kong-Macao Joint Laboratory was unveiled, and the number of records in Guangdong was refreshed. At the same time, Apple's iOS 26 system was officially launched, the Samsung AI Forum was held, Proofpoint and CrowdStrike released new solutions, and Germany launched consultation on the implementation of the AI bill. Achieve balanced reporting at home and abroad as a whole.

1. State Administration of Market Supervision: Conduct further antitrust investigation against NVIDIA

1. On September 15, it was confirmed that after a preliminary investigation, Nvidia violated the Anti Monopoly Law and its decision to impose conditions on the merger and acquisition of Melox.
2. The regulatory authorities decide to conduct further investigations in accordance with the law, involving market competition order and M&A compliance performance.
3. According to regulations, a fine of up to 1%-10% of sales in China can be imposed, or affect the supply ecology of AI chips.

2. Version 2.0 of the "Artificial Intelligence Security Governance Framework" was officially released

1. It was unveiled at the main forum of the National Cybersecurity Publicity Week on September 15, and was led by the National Internet Emergency Center.
2. On the basis of version 1.0, new risk classification research has been added to improve risk classification and dynamic governance.
3. Provide a unified starting point for AI security governance and strengthen the enforceability of policies, industries and academia.

3. Release of artificial intelligence-enabled network security test results in 2025

1. 256 teams from 169 units signed up, and 224 teams finally participated, covering 7 major test scenarios.
2. The test uses Huawei AI server and established computing architecture, and the second phase is completed at Changan Communications.
3. The results show that AI is significantly empowered in all scenarios, promoting "promoting use through testing and promoting safety through use".

4. The Guangdong-Hong Kong-Macao Greater Bay Area Generative AI Security Development Joint Laboratory was unveiled

1. On September 15, it was officially established in the Hetao Shenzhen-Hong Kong Science and Technology Innovation Cooperation Zone, focusing on generative AI security research.
2. Focus on breakthroughs in model evaluation, content security and cross-border governance, and promote joint research and pilot mechanisms.

5. Guangdong has 92 generative AI services

1. As of September 15, Guangdong has registered a total of 92 AI services, including 46 general large models and 46 industry large models.
2. The application covers government affairs, education, industry and automobiles, and supports the "computing power coupon" policy to promote the implementation.

6. Apple's iOS 26 global push, AI functions are enhanced with the system

1. It was officially launched on September 15, bringing AI functions such as Liquid Glass design and real-time Live Translation.
2. Some functions are opened in stages, and gradually implemented due to different languages and regions.
3. Mark "Apple Intelligence" capabilities and deep integration with the system to strengthen device-side privacy.

7. Samsung AI Forum 2025 opens in South Korea

1. Held on September 15-16, it is the ninth edition, bringing together global scholars and industry experts.
2. Focus on displaying the progress of basic research and the direction of computing power, and disclose the company's latest AI research and development achievements.

8. Proofpoint released an intelligent solution for agent AI communication

1. Released on September 15, it is known as the industry's first agent AI platform for compliance and risk.
2. Integrate with its DCG product line, and plan to launch more LLM functions in Q4/2025 and expand to ITM systems in Q1/2026.
3. It is mainly used for compliance monitoring and risk prevention and control of email and collaboration platforms.

9. CrowdStrike and Meta launch CyberSOCEval, a security AI evaluation benchmark

1. Launched on September 15, it will evaluate LLM performance in real scenarios for SOC (Security Operations Center).
2. Based on Meta CyberSecEval and CrowdStrike intelligence, the evaluation criteria are formed.
3. Establish comparative indicators for "AI + safe operation" to reduce the cost of enterprise selection.

10. Germany launches consultation on the implementation of the AI bill

1. On September 15, the German government issued a consultation for comments, planning the Federal Cyber Agency to serve as the regulator.
2. Emphasize the promotion of innovation and simplified regulation under the unified framework of the EU.
3. Clarify the compliance path for operating high-risk AI systems in Germany.

Frequently Asked Questions (Q&A)

Q: What is the potential impact of the Nvidia antitrust investigation?

A: The maximum fine is 1%-10% of the sales in China in the previous year, which may exacerbate the uncertainty of AI chip supply and affect industry cooperation and market expectations.

Q: What are the new points of the AI Security Governance Framework 2.0?

A: On the basis of risk classification, new risk classification is added, dynamic governance is emphasized, and the connection between standards and applications is strengthened.

Q: What are the new AI features in iOS 26?

A: Including AI capabilities such as Liquid Glass interface and real-time translation, some functions will be opened in phases depending on region and language.

Q: What is the industry value of CyberSOCEval?

A: Provide a unified evaluation benchmark to help enterprises compare the effectiveness of different LLMs in SOC scenarios such as threat detection and alarm handling.