Good afternoon, everyone. I am honored to be here to share some of my thoughts and experiences regarding case handling. My name is Leon Zhang and I am a lawyer from Kinding Law Firm, which focuses on Internet legal issues.
The topic I want to report on today is the responsibility tiering of large model service providers. Why is this a topic worth discussing? I would like to explain it from three aspects. First, I will briefly review the current state of responsibility tiering for online service providers.
Second, lets see this problem from an industry and business perspective, to know different roles which played by large model providers.
Finally, I will present what I believe to be a suitable framework for the responsibility tiering of large model service providers.
as we all know, China has largely inherited the DMCA’s safe harbor rules, which apply to various types of online service providers in the country, The requirement is that platforms must promptly remove infringing content upon receiving a complaint; otherwise, they bear joint liability with the direct infringer.
Initially, this demand applied to copyright-related infringements, but it gradually expanded to trademark \patent and almost all kinds of infringements,like reputation and privacy related infringement. This makes safe habor more likely a huge cost to internet sverce providers because the number of complaints is numerous.
Recently years, this trend has begun to change, especially with cases like the mini-program case and the cloud computing case that I represented. Courts in some cases have started applying different rules to these foundation service provider, like thransfer the complaits to direct infringer instead of remove infringe content.
After the implementation of the Civil Code, Article 1195 directly clarified that online service providers can take necessary measures based on their service types. Therefore, it can be said that we have made significant progress in tiering responsibilities and duties of care for online service providers.
With the emergence of large models, a similar issue arises. From an industry perspective, large models can be categorized into at least three different types: foundational models, industry-specific models, and application-layer models. These three models resemble a pyramid, structured from the bottom up. Generally, the models at the top are built upon those below. In this context, if the content generated by a model results in infringement, how should their responsibilities and duties of care be delineated? I believe that we should not use a single standard for judgment.
The Ultraman case involves this issue, where the defendants argue that their AI service is built on a general foundational model, and they have only made some targeted optimizations. The question arises: should this foundational model bear responsibility or prior duties of care? I propose a tiered responsibility framework to address this issue.
Firstly, for foundational models, there are three main duties of care that can exempt them from liability:
Public Law Duties: This includes obligations such as blocking information related to pornography, gambling, and drugs.
User Data Security Obligations: Particularly concerning the protection of personal information.
Compliance with Recognized Technological Ethics: This includes avoiding the generation of content that is clearly discriminatory, such as gender bias.
For industry-specific models, the focus should be on responding to regulatory requirements in that particular industry. For example, in the finance field, there are requirements for risk warnings regarding consumer investment behaviors.
Finally, for application-layer model service providers that directly offer services to specific users, they should take necessary measures against infringing information generated by the model that targets third-party rights holders.
Overall, the further down the tiers we look, the more the models can claim technical neutrality, while the higher up we go, the greater and more specific the duties of care become. This aligns with their technological characteristics and business revenue models.
That’s all I wanted to share on this topic. By the way, I am currently representing several other AI-related litigation cases, Many of these cases raise interesting issues, and I look forward to more opportunities to exchange ideas with all of you experts.
