Midwinter begins, and everything is renewed. With the care and support of many experts in the field of artificial intelligence intellectual property, IPRs, and lawyers, the 3rd Intellectual Property Frontier Artificial Intelligence Forum (IFAF 2023) was successfully concluded on December 8, 2023 at Four Points by Sheraton Beijing Haidian Yongtai Hotel.
The event was co-hosted by YIP Events & IP Frontier New Media & Compliance Plus, with the theme of "Intellectual Property Protection and Innovation Value in the Intelligent Era" during the two-day conference and half-day pre-conference seminar.
At the forum on December 8, Ling Zhaohua, head of intellectual property of Shuidi Company, delivered a keynote speech on "AIGC Era, Enterprise IP Compliance Practice and Exploration" at the conference. IP Frontier has compiled the content of Mr. Ling's on-site keynote speech into a text for the reference and Xi of intellectual property professionals.
table of contents
Part 1: IP compliance requirements and risk response for AIGC service providers
1.1 Requirements for AIGC enterprises in the Interim Measures for the Administration of Generative AI Services (the "Interim Measures").
1.2 Copyright compliance risks faced by AIGC enterprises
1.3 Copyright compliance suggestions for AIGC enterprises
1.4 Trademark compliance risks faced by AIGC enterprises
1.5 Trademark compliance advice for AIGC enterprises
1.6 Patent compliance risks and suggestions for AIGC enterprises
Part 2: IP compliance requirements and risk response for AIGC service users
2.1 Copyright compliance risks and suggestions for AIGC service users
2.2 AIGC generates copyright compliance risk cases for content users
2.3 Other IP compliance risks and suggestions for AIGC service users
Part 3: Some Thoughts and Suggestions on AIGC's Intellectual Property Compliance Governance
3.1 Thoughts on intellectual property compliance governance of AIGC training data
3.2 Reflections on the IP Compliance Governance of AIGC-Generated Content
Part 1: IP compliance requirements and risk response for AIGC service providers
1.1 Requirements for AIGC enterprises in the Interim Measures for the Administration of Generative AI Services (the "Interim Measures").
The basic requirement of AIGC enterprises under the Interim Measures is to respect the intellectual property rights of others. Article 4 stipulates that "the provision and use of generative AI services shall comply with laws and administrative regulations, respect social morality and ethics, and comply with the following provisions:...... 1. Respect intellectual property rights and business ethics, keep trade secrets, and do not use the advantages of algorithms, data, platforms, etc., to carry out monopoly and unfair competition; 2. Respect the legitimate rights and interests of others, and shall not endanger the physical and mental health of others, and shall not infringe upon the rights and interests of others' portrait, reputation, honor, privacy and personal information.......;
Article 7 of the Interim Measures specifically requires AIGC enterprises in the model training stage that reads: "Generative AI service providers (hereinafter referred to as "providers") shall carry out training data processing activities such as pre-training and optimized training in accordance with the law, and comply with the following provisions: 1. Use data and basic models from lawful sources, 2. Where intellectual property rights are involved, they shall not infringe upon the intellectual property rights enjoyed by others in accordance with the law; 3. Where personal information is involved, the consent of the individual shall be obtained or other circumstances provided for by laws and administrative regulations; ......”
Article 9 of the Interim Measures provides that "the provider shall bear the responsibility of the online information content producer in accordance with the law and perform the obligations of online information security." Where personal information is involved, bear the responsibility of personal information handlers in accordance with law, and perform personal information protection obligations. Providers should sign service agreements with users of generative AI services who register their services (hereinafter referred to as "users"), clarifying the rights and obligations of both parties. ”
1.2 Copyright compliance risks faced by AIGC enterprises
1.2.1 AIGC enterprises are classified as copyright risks
The stages of copyright risk for AIGC enterprises are still mainly in the stage of model training and content generation. The model training stage is the risk brought by the training data and the underlying model, such as the risk of infringement of the reproduction right of the work, the risk of violating the CC license of the work, the risk of violating the open source license of the code, and the risk of violating the open source model license agreement. In the content production stage, the copyright compliance risks of AI-generated products are mainly used, such as the risk that the generated products infringe the right of reproduction, adaptation, compilation, translation, and information network dissemination.
1.2.2 Common datasets for AIGC large model training (data source)
Common datasets (data sources) for AIGC large model training and their functions are as follows:
1. CommonCrawl (free commercial, abide by ToU): The super-large dataset crawled by the website from 2008 to the present, including the original web page (web page);
2. Wikipedia (in compliance with the CC-BY-SA-4.0 license): Wikipedia website dataset, encyclopedia knowledge content covers more than 20 languages around the world (encyclopedia);
3. Github (abide by the open source license): GitHub public datasets, only keep the projects (code classes) released under the Apache, BSD and MIT licenses;
4. Gutenberg & Books3 (comply with ToU): Gutenberg is a public domain book, and Books3 is also a public dataset of books (books);
5. ArXiv (CC license): a free online preprint dataset of papers, including more than 2.4 million academic papers, non-peer-reviewed journals (papers);
6. Stack Exchange (in compliance with CC-BY-SA-4.0 license): questions and answers compiled and edited by the Stack Overflow Q&A website (Q&A);
7. LAION-5B (CC-BY-4.0 Compliant): A graphic dataset (image class) containing 5.85 billion image-text pairs filtered by the CLIP model.
1.2.3 Copyright risk cases of well-known foreign AIGC enterprises
When AIGC trains large models, the training data may be considered to infringe copyright, and there are few cases in China (there have been similar disputes in Xiaohongshu recently, which are still in the stage of filing), and there are many more in the United States. The typical cases are as follows, and there is still no judgment yet.
1.2.4 "Fair Use" Determination
(1) Differences between China and the United States in the determination of "fair use".
When sued as an AIGC company, it may use the fair use defense, i.e., whether the use of copyrighted works to train a model constitutes "fair use". The "Chinese standard" is the "three-step test method" and "closed enumeration + semi-exhaustion": the use of a work under the following circumstances (specified special circumstances) may be carried out without the permission of the copyright owner and without payment of remuneration, provided that the name or title of the author and the title of the work and the title of the work shall be indicated, and shall not affect the normal use of the work, and shall not reasonably damage the legitimate rights and interests of the copyright owner: (1) for personal Xi study, research or appreciation...... ...... (13) Other circumstances provided for by laws and administrative regulations.
The "American Standard" is the "four-factor analysis" and open-ended enumeration: the use of a work for the purposes of criticism, commentary, journalism, teaching, scholarship, or research,...... , which is a fair use ....... When determining whether the use of a work constitutes fair use, the factors to be considered should include:
1. The nature and purpose of the use of the work, which is the factor with the highest weight in the judgment of the U.S. court;
2. The nature of the work being used;
3. The quantity and quality of the parts to be used;
4. The impact of use on the potential market or value of the work.
and (2) the reform of the "fair use" standard by Chinese courts
Since the Copyright Law of the People's Republic of China provides exhaustive provisions on the restriction of rights, and the types are too few to meet practical needs, the Supreme People's Court has issued a judicial policy to allow the courts to break through the express limitations of the Copyright Law when adjudicating copyright infringement disputes, and determine that an unauthorized use of a work in a case does not constitute infringement. Article 8 of the Opinions on Several Issues Concerning Giving Full Play to the Role of Intellectual Property Adjudication Functions in Promoting the Great Development and Prosperity of Socialist Culture and Promoting the Independent and Coordinated Development of the Economy promulgated by the Supreme People's Court in 2011 points out that "the proper use of copyright limitations and exceptions shall be used to correctly determine the legality of the alleged infringement,......。 In special circumstances where it is truly necessary to promote technological innovation and commercial development, considering factors such as the nature and purpose of the use of the work, the nature of the work being used, the quantity and quality of the part being used, and the impact of the use on the potential market or value of the work, if the use does not conflict with the normal use of the work and does not unreasonably harm the legitimate interests of the author, it may be found to be fair use. ......”
(3) Google Digital Library Case: Similar Cases in China and the United States Are Not Judged (Transformative Use?)
The case was prosecuted in both China and the United States, and the U.S. court directly determined that Google's copy-scanning behavior was highly transformative through the first element of the four-element law, and that the number of words displayed was limited through the third element, and finally found that Google constituted fair use. However, the Chinese court held that Google's act of disseminating information on the network of the work was fair use, but its reproduction constituted infringement, and the overall standard for determining fair use was higher.
Specifically, the Beijing Higher People's Court held in the second instance ((2013) Gao Min Zhong Zi No. 1221) that the determination of fair use outside the specific circumstances provided for in Article 22 of the Copyright Law should be strictly controlled. Unless the user fully proves that the use constitutes fair use, the use shall be presumed to constitute infringement. In determining whether fair use is constituted, factors such as the purpose and nature of the use of the work, the nature of the copyrighted work, the quality of the part used, its proportion in the entire work, and the impact of the use on the actual and potential market and value of the work should be considered. The burden of proof shall be borne by the user for the factual issues involved in the above considerations. In this case, Google only submitted evidence to prove that the Chinese court had no jurisdiction over the case, and did not submit evidence on whether the copying constituted fair use, so its claim that the copying constituted fair use was insufficient.
The U.S. Court of Justice for the Second Circuit (The Authors Guild v. Google, Inc.) held that Google's unauthorized electronic scanning of copyrighted books, setting up search functions, and displaying fragments of the books on the Internet are non-infringing fair uses. Google's scanning is highly transformative, the amount of text displayed is limited, and the fragments made available to the public do not compete with or substitute for the original work. Therefore, even if Google is a profit-seeking commercial company, this does not prevent the actions of the Google Digital Library from being considered fair use.
(4) Does AI training constitute transformative use?
Transformative use was first proposed in 1990 by U.S. Judge Pierre in the article "Fair Use Standard", and used it as a method to evaluate the first element of fair use (the purpose and characteristics of the use of the work), pointing out that the court should evaluate whether and to what extent the new work is a transformative use of the original work, and this evaluation standard was later adopted by the U.S. Supreme Court.
Professor Wang Qian's interpretation of the criteria for determining whether it constitutes transformative use is as follows:
1. The use of the original work is not for the purpose of simply reproducing the literary or artistic value of the original work itself or realizing its intrinsic function or purpose;
2. By adding new aesthetic content, new perspectives, new concepts or through other means, the original work has a new value, function or nature in the process of being used;
3. The original function or purpose of the work has been changed.
1.3 Copyright compliance suggestions for AIGC enterprises
(1) Model training stage
In view of the copyright compliance suggestions in the training stage of AIGC enterprise models, firstly, it is recommended to give priority to the use of works and data that have entered the public domain and are not protected by the Copyright Law, including works that are not within the meaning of the Copyright Law, works that have expired the protection period, and works in which the author declares that the copyright has been permanently waived.
Second, you can use open source models, open source code, open source datasets, and works under the CC license, but you should try to choose a loose open source license and a friendly CC license, and abide by the terms of the license agreement.
In addition, the training data is purchased through a third party, and the supplier is required to provide a flawless guarantee or non-infringement guarantee for the intellectual property rights of the training data through agreements and other means, and stipulate a clause for the transfer of infringement liability.
Thirdly, standardize the use of crawlers, Open APIs and other technical means to obtain training data, focusing on evaluating the compliance of crawler behavior and whether the data scraping behavior has damaged the technical protection measures of the website.
Secondly, it is advisable to be cautious about the use of works and data with strong copyright, and if it is really necessary to use them, it is recommended to obtain legal and effective authorization from the right holder in advance, and use them in compliance within the scope of authorization.
Finally, an effective "opt-out mechanism" should be established to technically facilitate the search and retrieval mechanism for copyright owners of works, and allow copyright owners to freely choose whether to delete their copyrighted works from the training database.
(2) Content generation stage
First, establish a back-end sensitive thesaurus to include prompts that may lead to a high risk of infringement in the database, and take technical measures to identify, review and block high-risk content entered by users.
Second, copyright filtering technology, image similarity detection technology and other technical means are used to identify, compare, filter and regenerate AI-generated high-risk content.
Third, add clauses such as the agreement on the ownership of rights, the assumption of liability for infringement, and the restriction of use of the generated content in the service agreement, user agreement, and other relevant documents.
Finally, establish feedback channels for content infringement complaints, and promptly take measures such as disconnecting links, deleting, and notifying users of the identified infringing content.
1.4 Trademark compliance risks faced by AIGC enterprises
1.4.1 Common types and scenarios of trademark compliance risks for AIGC enterprises
(1) Category 1 - risk of infringement of trademark use
The risks of trademark infringement are mainly as follows:
1. Infringement of the name and trademark of various AIGC products such as large models, chatbots, AI assistants, etc.;
2. Trademark infringement of names/nicknames/avatars/icons of apps, mini programs, public accounts, etc.;
3. Trademark infringement of the name of the self-made column, the name of the program, the name of the live broadcast room, and the domain name of the official website;
4. Highlight the use of trademark infringement such as enterprise name, company abbreviation, organization name, etc.
(2) The second category - the risk of general trademark violations
The risks of trademark infringement are mainly as follows:
1. Use the logo that is not allowed to be used as a trademark in the product name/brand name according to the law;
2. Use unregistered trademarks as registered trademarks, such as marking registered marks without authorization;
3. Change the logo of the registered trademark or use it beyond the scope of the approved goods of the registered trademark;
4. Preemptively registering hot spots, public events, trademarks with certain influence on others, and malicious registration of trademarks.
(3) Category 3 - risk of non-registrability of trademarks
Trademarks may be at risk of non-registrability:
1. Lack of distinctiveness (generic name, model, graphics; only indicates service content, characteristics, etc.);
2. Deceptive (misidentification of service content, quality, source, etc.; misidentification of names and characters);
3. Adverse effects (moral, political, economic, cultural, ethnic, religious, social, etc.);
4. Prior rights (copyright, design patent right, trade name right, name right, portrait right, landmark, etc.).
1.4.2 Typical trademark compliance risk cases of AIGC enterprises
In August 2023, OpenAI sued Open Artificial Intelligence and its president, Guy Ravine, with the main complaint asking the court to determine that the two defendants constituted trademark infringement and unfair competition, constituted fraudulent registration, requested the court to cancel the registration of the defendant's infringing trademark "Open AI", and transferred the domain name registered by the defendant to the plaintiff www.open.ai. However, so far, OpenAI has not successfully registered the OpenAI trademark in the United States, while the defendant Open Artificial Intelligence has successfully registered the Open AI trademark, and the filing date of the Open AI trademark registered by the defendant is December 11, 2015, which happens to be the date when OpenAI officially announced its establishment, and this news has been reported in the New York Times This is also the reason why OpenAI accused the defendant of fraudulent registration. In addition, the defendant's Open AI trademark was registered on the U.S. Supplementary Trademark Register, indicating that the USPTO also believed that the trademark lacked distinctiveness and could not be directly registered in the main register. Compared with a registered trademark in the main register, a trademark registered on the supplementary register does not need to go through an opposition period and its rights are relatively unstable, but it can also prevent the registration of a later identical or similar trademark, and can also legally protect its rights based on the trademark rights. In short, there are still many legal issues worth digging out and discussing in this case, and it is difficult for us to predict the final direction of this case. But for both OpenAI and the defendant, perhaps settling the OR acquisition is the best win-win solution!
In addition, there is also a risk of trademark registrability for AIGC product names. For example, it is doubtful whether the trademarks in the red boxes below are distinctive.
For another example, many companies use "COPILOT" as a product name in the field of artificial intelligence, so the distinctiveness of using "COPILOT" as a trademark registration alone is very low.
1.5 Trademark compliance advice for AIGC enterprises
The trademark compliance suggestions for AIGC enterprises mainly include the following points:
1. Enterprises should develop the good awareness Xi of "register first, use later" of trademarks, and advance trademark brand planning to the early stage of AI project/product development;
2. "Professional matters should be handed over to professional people" - professional advice should be listened to for brand naming and trademark protection related to intellectual property legal affairs or external cooperative service agencies;
3. The product/brand name must undergo a professional assessment before it is put into use, including trademark infringement risk assessment, trademark general violation risk assessment and trademark registrability assessment;
4. After the trademark is registered, it is necessary to standardize the use, and the use of the trademark logo should be avoided as much as possible, especially not to change the use of the main identification part of the trademark, which not only has the risk of trademark general illegality, but also the risk of infringing on the exclusive right to use the trademark of others;
5. Stay away from trademark squatting (hot spots, preemptive registration of others, etc.), malicious trademark registration and trademark hoarding without the purpose of use.
1.6 Patent compliance risks and suggestions for AIGC enterprises
The patent compliance risks of AIGC enterprises can be divided into the following 14 risk scenarios, with a total of 33 risk points.
It can be suggested that it is necessary to pay close attention to the AI patent application and authorization of domestic and foreign friends, make FTO for self-developed large models and products in advance, and assess the risk of patent infringement; in the process of cooperative development, it is necessary to agree on the ownership and use of prospective intellectual property rights, and agree on the exemption clause and the way to bear the liability for patent infringement; carefully evaluate the intellectual property protection methods of AI models and their application products (open source or closed-source, whether the core patent is an invention patent or a trade secret?); The examination policy of model-related patents should be formulated in accordance with local rules, and the abnormal patent application behavior should be avoided that is not for the purpose of protecting innovation and is not based on real inventive and creative activities.
The second part is the IP compliance requirements and risk response of AIGC service "users".
2.1 Copyright compliance risks and suggestions for AIGC service users
2.1.1 Copyright risk scenarios for AIGC service users
The main copyright risk scenarios for using AIGC products are as follows:
(1) Use AIGC products to generate content that is substantially similar to the copyrighted works of others, and commercialize them;
(2) Directly input the copyrighted works of others for modification, adaptation or polishing of AIGC products;
(3) Using pirated AIGC products or using AIGC products through unofficial authorized accounts/API interfaces;
(4) Failure to comply with the ToU or open source license of AIGC products when using them, or use them beyond the scope of their authorization.
2.1.2 Advice on copyright risk compliance for AIGC service users
For the above risk scenarios, we recommend:
(1) After AIGC generates the suspected infringing content, it can carry out AI fine-tuning, that is, regenerate the content by adjusting the prompt/parameters, etc., or directly manually adjust and modify the generated content;
(2) Avoid directly importing other people's copyrighted works or using Prompt to let AIGC products export others' copyrighted works (subject to company regulations);
(3) Obtain AIGC products and accounts through official channels or legal channels;
(4) When using AIGC, it strictly abides by its service agreement or its applicable open source license, and does not exceed the scope and purpose of its authorized use.
2.2 AIGC generates copyright compliance risk cases for content users
1. The first case of copyright infringement of pictures generated by AIGC in China [(2023) Jing 0491 Min Chu No. 11279]
The basic timeline of this case is as follows:
- On February 24, 2023, the plaintiff used the AI software Stable Diffusion to generate the AI picture involved in the case by entering prompt words, and posted it on the Xiaohongshu platform on February 26, with labels such as "AI illustration".
- On March 2, the defendant published an original poem on the Baijiahao platform, in which the AI picture involved in the case was used as one of the accompanying pictures without the plaintiff's permission, and the plaintiff's signature watermark on the Xiaohongshu platform was intercepted.
- On May 25, the Beijing Internet Court formally filed a case against the plaintiff against the defendant for infringement of the right of authorship of the work and the right of information network transmission.
- On August 24, the court held a public hearing of the case and broadcast the trial process live, which attracted wide attention in the industry.
- On November 27, the court of first instance found that the defendant had committed infringement, and ordered the defendant to apologize and compensate the plaintiff for economic losses of 500 yuan.
The core view of the Beijing Internet Court in the first instance held that:
(1) From the perspective of the generation process of the pictures involved in the case, the plaintiff made a certain amount of intellectual investment, such as designing the presentation of the characters, selecting the prompt words, arranging the order of the prompt words, setting the relevant parameters, selecting which picture meets expectations, etc., so the pictures involved in the case have the element of "intellectual achievement";
(2) The plaintiff designed the picture elements such as the characters and their presentation methods through prompts, and set the layout and composition of the picture through parameters, reflecting the plaintiff's choice and arrangement;
(3) When people use artificial intelligence models to generate pictures, they are still essentially people using tools to create, that is, it is people who invest in intelligence rather than artificial intelligence models in the entire creative process, and as long as it can reflect people's original intellectual input, it should be recognized as works and protected by copyright law.
2. Tencent v. Yingxun Case [(2019) Yue 0305 Min Chu No. 14010]
The People's Court of Nanshan District, Shenzhen, Guangdong Province, held that:
(1) The article involved in the case was generated by the plaintiff's main creative team using Dreamwriter software, and its external performance met the formal requirements of the written work,......。
(2) To sum up, from the analysis of the external form of expression and the generation process of the article involved in the case, the specific form of expression of the article and its origins from the creator's personalized selection and arrangement, and the creative process technically "generated" by the Dreamwriter software all meet the protection conditions of the copyright law for literary works, and this court finds that the article involved in the case is a literary work protected by the mainland copyright law. This court found that the article in question was a legal person work created by the plaintiff.
3. "Fei Lin v. Baidu" [(2019) Jing 73 Min Zhong No. 2030]
The Beijing Intellectual Property Court held in the second instance that:
The text content of the article involved in the case was not automatically generated by the "visualization" function of Wolters Kluwer's advanced library, but was independently created by Film Law Firm, which is original and constitutes a written work. Although the graphics in the article involved in the case are produced by the film law firm based on the collected data and the use of relevant software, although they will show different shapes due to data changes, the difference in the shape of the graphics is based on the data differences, not based on the creation, which cannot reflect the original expression of the film law firm. Therefore, the graphics in the article involved in the case did not constitute graphic works, and the court of first instance found that this was correct, and this court affirmed it.
2.3 Other IP compliance risks and suggestions for AIGC service users
2.3.1 Other IP risk scenarios for AIGC service users
Other risk scenarios for AIGC service users are generally as follows:
(1) The risk of leakage of the company's trade secrets and business data caused by the input of improper information;
(2) the risk of false commercial promotion due to the improper generation and use of false content;
(3) the risk of unfair competition is posed due to the improper use of the packaging/decoration of others' well-known goods;
(4) The risk of trademark infringement caused by the improper use of content containing the trademarks of others.
2.3.2 Other IP compliance suggestions for AIGC service users
For the above risk scenarios, we recommend:
(1) It is recommended that the company formulate corresponding management systems, norms or guidelines for employees' use of AIGC products, and prohibit employees' improper use;
(2) Suggest that the company take specific technical measures to prevent employees from entering inappropriate information such as trade secrets, business data, and personal information in AIGC products, if conditions permit;
(3) It is recommended that the Company take inappropriate content filtering measures when conditions permit, and filter and block the inappropriate content generated by AIGC in a timely manner to avoid improper use;
(4) It is recommended that the company's legal or intellectual property department do a good job of reviewing the content generated by AIGC before it is put into commercial use, so as to avoid the risky content being used for commercial promotion.
The third part is some thoughts and suggestions on AIGC's intellectual property compliance governance
3.1 Thoughts on intellectual property compliance governance of AIGC training data
From an objective point of view, the large amount of data required for large model training, the variety of data types, and the scattered data sources lead to a large number and wide dispersion of intellectual property rights holders involved in the training data, and the traditional "authorization-licensing" model has low operability in the model training stage.
From the perspective of the mode and purpose of use, the use of the work as training data is obviously different from the traditional use of the work, and the purpose of use is not to reproduce the original beauty and function of the original work, but to create a new way of use and new value of the work.
From the perspective of use impact, the use of the work in the model training stage did not affect the normal use of the work, nor did it unreasonably damage the legitimate rights and interests of the copyright owner, nor did it create a competition or substitution relationship with the original work in the market.
Based on the above analysis, the following governance suggestions can be given: (1) amending the Copyright Law to add the use of works for AIGC model training as a special fair use case, but since the Copyright Law has just been revised, the possibility of revising it in the short term is very low; (2) interpreting AIGC model training as a fair use case by issuing judicial interpretations or judicial policies; (3) enacting special legislation on AI (such as the Artificial Intelligence Law (Model Law)) or enacting/ Revise administrative regulations (e.g., Regulations for the Implementation of the Copyright Law) to make AIGC model training a fair use situation, and (4) adopt other methods that can reduce the legal liability or burden of AIGC enterprises, such as the use of works first, payment later, centralized licensing of works (one-stop licensing), and the promulgation of licensing guidance documents specifically for the works involved in training data.
3.2 Reflections on the IP Compliance Governance of AIGC-Generated Content
From an objective point of view, the content generated by AIGC can be modified, adjusted, regenerated, deleted, discarded, etc. before it is officially "used", and there is no inevitable/unavoidable use of "infringement".
From the perspective of the mode and purpose of use, most of the current "users" use the content generated by AIGC in the same way and purpose as the traditional use of ordinary works, all of which are to reflect the original beauty and function of the work/content, and do not create new value, new beauty, new meaning, and new use.
From the perspective of the impact of use, the use of the "new work" in the content generation stage affects the normal use of the original work (the infringed work), has a competitive or substitution effect on the original work in the market, and unreasonably damages the legitimate rights and interests of the original copyright owner.
Based on the above analysis, the following governance suggestions can be given: (1) the AIGC-generated content can be assessed for infringement in accordance with the existing copyright infringement judgment principle of "contact + substantial similarity", (2) the use of AIGC-generated content should not be used as a fair use situation to avoid infringement liability, and (3) for the allocation of infringement liability between AIGC service providers and users, it is recommended that the agreement take precedence, and the "safe harbor principle" can be applied by reference.
Author: Ling Zhaohua
Edited by Eleven