
AI-generated summary
The U.S. technology community has recently been divided over whether the pace of cutting-edge AI development should be slowed down, and the balance between security and innovation has become a hot topic.
Huida CEO Huang Renxun and Anthropic CEO Amoday served as keynote speakers at the Salesforce Annual Conference today. Two people with different views on the development speed and regulatory stance of AI have appeared one after another. Huang Renxun said that innovation speed and product safety are not the only choices. Amoday advocated the establishment of common standards in the industry.
The U.S. technology community has recently been divided over whether the pace of cutting-edge AI development should be slowed down. At the Salesforce annual conference today, Anthropic CEO Dario Amodei and NVIDIA CEO Jen-Hsun Huang were asked about related issues when they talked with Salesforce CEO Marc Benioff.
Talking about AI safety, Armoday, who was first on the stage, used the analogy of safety incidents in automobile companies and pointed out that when safety incidents occur in the industry, the responsible approach is not to attack competitors, but to first review and improve their own measures, strengthen transparency and safety investment, then call on the industry to jointly set higher standards, and finally promote coordination and regulation at the international level.
Amoday also said that what surprised him most in the past 10 years was not the improvement in AI model capabilities, but the speed of expansion of AI companies, products and commercial applications. He said that technology will continue to advance, and even if technology stops advancing from now on, companies may currently only realize 5% to 10% of their potential value. This shows that there is still huge room for popularization and application of existing AI tools.
Salesforce launched Koa, a dedicated customer relationship management (CRM) model built in partnership with Huida, at the conference today.
Huang Renxun, CEO of Huida, took the stage in person and made his final appearance. He was asked what he thought of the debate on the pace of AI research and development, safety and the future of AI models. Safety is critical in many ways, but "safety is an engineering problem," he said.
He said that it is necessary to ensure that the industry builds testing environments in a safe manner and also tests products. "If you build a product or service and you are not confident in its functionality, capabilities or security, then don't launch it."
He emphasized that companies should adjust their pace until they are confident in releasing results that the market appreciates. "The market mechanism has already existed. We don't need any new laws, we don't need new regulations. We just need each company to make its own decision."
Huang Renxun said that he does not believe that the pursuit of innovation speed must sacrifice product safety. This is not a matter of choosing one or the other. "You can definitely take into account both." Run as fast as you can, he said, but if at any time you feel that the company has lost control or that the product may be unsafe, take a pause and make sure the problem is dealt with properly.
AI outlook — possibilities, not facts
Major AI companies to announce voluntary safety standards or transparency reporting mechanisms in the coming months
Likely · Within months
The Koa model of Huida and Salesforce will be gradually introduced to enterprise customers for trial in the next quarter.
Possible · Within months
The 2026 National Cyber Security Awareness Week was held in Jinan. Experts pointed out that AI large models have caused a sharp increase in the number of network vulnerability discoveries, attacks have industrialized production capabilities, and the risk level has fundamentally changed. Representatives from Huawei, China Southern Power Grid and the Ministry of Public Security called for a shift from passive defense to active coordinated defense and digital identity protection to build a three-dimensional defense system.

After Apple released iOS 27, some users reported reduced battery life and body heat. Apple officially explained that this was a temporary phenomenon caused by background programs re-indexing and downloading assets after a system update. It was recommended that users wait a few days for the device to complete settings. Battery performance may even improve in the future.

Sony was found to have quietly changed the description of the number of operating system updates on its Xperia 1 VIII and Xperia 10 VIII overseas official specification pages from "4 times" to "up to 4 times" without prior announcement, which aroused the attention of overseas users. Both the Japanese domestic version and the Taiwan product page have long used the expression "up to 4 times" and there has been no change. This move may affect consumers' confidence in Sony's software support commitment, especially considering that the Xperia 10 VI has experienced a similar lack of support.

Artificial intelligence is gradually integrating into daily life from a tool that passively answers questions. AI Agents actively perform tasks such as calendar organization, traffic information, smart home control, etc., making life more convenient. However, it also accumulates personal data, raising privacy concerns.

The Apple iPhone 18 Pro series will go on sale on Friday, September 18. Chunghwa Telecom, Taiwan Mobile and Far EasTone Telecommunications will simultaneously hold first sales events, offering discounts, lucky draws and gifts. Chunghwa Telecom will use a pre-purchase lottery system to issue shopping bonuses and lucky prizes. Taiwan Mobile invites goddesses to support and provide project application gifts, while Far EasTone Telecommunications provides queue discounts and old phone trade-in discounts.

The article points out that low-quality false information ("digital swill") generated by AI is polluting the network ecology. It recommends using AI to govern AI, building high-quality corpora, optimizing algorithm recommendations, and improving public information literacy to make real and valuable information standard for AI content.