21st century business herald reporter Xiao Xiao reports from Beijing.
This week (8.12-8.17), overseas anti-monopoly, domestic Internet content governance, and artificial intelligence all witnessed remarkable compliance incidents.
Anti-monopoly: the General Administration of Market Supervision is speeding up the formulation of the national standard of "Regulations on the Management of Fair Competition and Compliance for Operators";
Personal information security: home appliance companies () responded to the question of privacy disclosure, and smart home appliances lacked regulatory rules;
Internet content governance: when the "sports rice circle" was rectified, there were five administrative penalties;
In terms of artificial intelligence: after receiving the "lawyer’s letter warning" from HowNet, Secret Tower AI will no longer include HowNet; Limit the proliferation of crawler websites;
Overseas: a startup company uses algorithms to solve the problem of AI plagiarism; Copilot built into Microsoft system was exposed to data leakage; After Google lost the antitrust case, it was exposed to consider splitting the business; Universal Music and Meta expand cooperation and negotiate AI music copyright issues.
I. Anti-monopoly
1. The General Administration of Market Supervision is speeding up the formulation of the national standard "Standards for the Management of Fair Competition and Compliance of Operators".
On August 16th, the State Council held a series of press conferences on the theme of "Promoting High-quality Development". At the meeting, the State Administration of Market Supervision revealed that this year, the anti-monopoly normalization supervision in the fields of digital economy was further promoted, and platform enterprise operators were unconditionally approved to report 48 cases in a centralized manner according to law. At present, we are speeding up the formulation of the "Regulations on the Management of Fair Competition and Compliance for Operators".
Second, personal information security
1. Cobos responded to the question of privacy disclosure, and smart home appliances lacked regulatory rules.
Recently, two security researchers attended the Def Con security conference, saying that they found that there were security problems in Cobos’s sweeping robot products. After connecting the Cobos robot via Bluetooth, hackers could remotely control the product through its own WiFi connection function, and access the room map, camera, microphone and other functions and information in its operating system.
In response to the above problems, Cobos said in an interview with Southern Finance and Economics all-media reporter that data security and user privacy are one of the most important issues in Cobos. The Cobos Robotics Safety Committee has made internal research and comments on products in network connection, data storage and other issues, and its conclusion is that the probability of these security risks in users’ daily use environment is extremely low, which requires professional hacking tools and close contact with machines, so users don’t have to worry about it. Nevertheless, Cobos will actively optimize its products based on research and appraisal findings.
Nancai Comments: Behind the frequent safety problems of all kinds of smart home appliances, on the one hand, the status quo of enterprise safety construction needs to be further improved, on the other hand, there are also cases of lack of regulatory rules in related fields. At present, although China has laws and regulations on network security, hardware design and manufacturing, there has been a lack of corresponding subdivision standards in the field of intelligent products combining software and hardware, and there is no way to talk about all kinds of security requirements and safeguard measures extended on this basis.
Third, Internet content governance
1, "sports rice circle" rectification, administrative punishment has reached 5.
After the women’s table tennis singles final of the Paris Olympic Games, four platforms, Weibo, Tik Tok, Today’s Headline and Aauto Quicker, released reports on the management of "uncivilized watching" during the Olympic Games. Tik Tok and Aauto Quicker directly used the word "rice circle" in the headlines to describe these chaos. There are no fewer than 20,000 articles judged as inappropriate comments on each platform, so hundreds of accounts are banned. Last week, Weibo opened a "rice circle complaint area" for this purpose.
Public security organs are also involved in governance. On August 6, Beijing Daxing Public Security Bureau found that there was a fan who lied about the relationship between athletes and coaches, and the fan had been administratively detained. On August 15th, the Ministry of Public Security once again announced four typical cases of cracking down on and rectifying crimes involving "rice circles" in the sports field, and public security organs in Guangdong, Shandong, Hebei and Henan all took compulsory administrative punishment.
Nancai Comments: Fanquan is an organization with clear different discourse rights and upper and lower power levels, which can have a great influence. According to the scholars interviewed, rice circles cannot be simplified as a gathering of fans, but it is not a natural problem. Rice circles are the product of the appeal of sports industry, platform technology and entertainment capital. Only by recognizing this can we talk about governance issues.
Fourth, artificial intelligence
1. Restrict the proliferation of crawler websites, and the contradiction between data supply and demand is heated up.
Recently, Andrew Ng, an artificial intelligence expert, mentioned a study on data licensing on the website The Batch. The study found that various websites crawled by C4, RefineWeb, Dolma and other open source data sets are rapidly tightening their licensing agreements. Since the appearance of GPTBot (in mid-2023), the number of websites that are completely restricted at the level of robots.txt has surged. OpenAI, Anthropic and Common Crawl rank in the top three, all of which reach more than 80%, but website owners are usually tolerant and open to non-AI crawlers such as Internet Archive or Google Search.
Researchers are worried that this will not only affect the training of commercial AI models, but also hinder the research of academic circles and non-profit organizations.
2. After receiving the "lawyer’s letter warning" from HowNet, Secret Tower AI will no longer be included in HowNet.
Recently, HowNet accused "Secret Tower AI Search, Secret Tower AI Search APP to provide users with our academic literature titles and abstract data, and the number is huge. This behavior has seriously violated the legitimate rights and interests of our company without our permission. " HowNet requires that China HowNet doesn’t want to be searched by Secret Tower Technology, and requires Secret Tower AI to immediately disconnect the search results.
On August 16th, Mi Ta AI sent a message in response to HowNet, stressing that he only included the abstract and title of the paper, but did not include the content of the article itself. Reading the paper still needs to jump to the website to get it. Secret Tower AI expresses that it does not understand but respects the choice of HowNet. From now on, it will no longer include the titles and abstract data of HowNet literature, but instead include the data of other authoritative knowledge bases in Chinese and English.
Nancai comment: HowNet’s accusation mainly focuses on two contents: document title and abstract. Simple titles and catalogues are generally not considered to constitute works (in the sense of copyright), nor do they involve infringement. However, the abstract part is likely to be recognized as a work, and unauthorized quotation may constitute infringement.
It is worth mentioning that at a forum attended by reporters on the 21st of last month, a senior executive of HowNet revealed that HowNet has recently begun to explore the way to realize data and has reached cooperation with Huawei.
V. Overseas
1. A startup company uses an algorithm to solve the problem of AI plagiarism.
On August 6th, the startup ProRata.AI was established in California, and its founder was Bill Gross, who had put forward bidding ranking advertisements and paid for advertisements on a pay-per-view basis. Now Bill Gross has proposed a business model in the AI era: AI pay-per-use. Relying on a patented algorithm of the company, the output of AI will be disassembled into different parts, the corresponding copyright source will be found, and the income will be distributed according to the output proportion.
Although there are no products that have fully landed, large content copyright owners such as Financial Times, Fortune and Universal Music Group have signed cooperation agreements with ProRata. At present, the company has received $25 million in Series A financing.
2. Copilot built into Microsoft system was exposed to data leakage.
A researcher at the Black Hat 2024 conference revealed an amazing discovery. Copilot, Microsoft’s AI assistant, has multiple security vulnerabilities, which can be used by attackers to steal sensitive data and even become a powerful phishing attack tool. The research results show that thousands of accessible artificial intelligence assistants can reveal sensitive data and company credentials that malicious actors can use.
In addition, the researchers demonstrated how attackers can exploit these vulnerabilities, and they can trick Copilot into modifying the information of bank transferors without obtaining an enterprise account, and even attack without obtaining the target employee to open the mail.
Nancai Comments: Copilot has two main problems: First, the training data will inevitably contain privatization information. When providing AI with data access rights, these data become the attack surface of prompt injection. To some extent, if a robot is useful, then it is fragile; If it is not fragile, it is useless. This is a fundamental contradiction; In addition, the interaction of public clouds also aggravates the risk.
In view of the security risks brought by the development of AI technology, Zhang Yaqin, academician of China Academy of Engineering and president of Tsinghua University Intelligent Industry Research Institute, suggested that enterprises or institutions engaged in cutting-edge large-scale models should invest 10-30% in related research or product development.
3. After Google lost the antitrust case, it was exposed to consider splitting the business.
On August 13th, local time, according to foreign media quoted people familiar with the matter, after the US court ruled last week that Google’s search business violated the US antitrust law, the winning US Department of Justice has taken the rare punishment of splitting Google into consideration.
According to informed sources, if the Ministry of Justice pushes forward the spin-off plan, Google is most likely to be spun off from Android operating system and web browser Chrome. Once the Google spin-off plan of the Ministry of Justice is implemented, it will become the largest spin-off event of an American company since AT&T was spun off in 1984.
4. Universal Music and Meta expand cooperation and negotiate AI music copyright issues.
On August 12th, Universal Music Group announced that it has expanded its multi-year cooperation agreement with Meta, which allows users to share songs from Universal Music Library on Meta’s platforms (Facebook, Instagram, Horizon, Threads and WhatsApp) without infringing copyright. The most striking thing about the new agreement is that it points out that the two companies are solving the problem of "unauthorized artificial intelligence generating content".
At the beginning of this year, Universal Music once "broke up" with TikTok. In a statement released in February, Universal Music emphasized that the company was worried that the AI technology of the platform threatened the artist’s copyright pool. The two companies reached a truce agreement in May, and Taylor Swift and other artists returned to TikTok for music rights.
