AI Search
The generative AI wave is transforming how cultural and entertainment content is created and produced. AI tools are now widely used in the making of text, music, art and audiovisual works.
It is also common to commission third-party production teams to create using AI tools, which carries compliance risks such as data leaks and intellectual property infringement.
A simple clause requiring the contractor to “bear all legal liability arising from the use of AI tools” falls short of the complex risks involved. Clients should establish a comprehensive legal risk prevention and control framework for AI applications.
Contracts should require contractors to use only secure, compliant AI tools with valid commercial licences. That protects clients from subsequent penalties, infringement claims or limits on use where a tool’s legality or rights status is flawed.
Contractors should, for example, warrant that the AI tool and its underlying foundation model have a lawful source, and should choose tools that have completed statutory algorithm filing, passed security assessment, and been officially published.
Overseas AI tools need particularly strict control, since a contractor’s improper use of AI tools or services that are not filed in China, or fail to meet Chinese regulatory standards, exposes the client to compliance risks.
Before using any AI tool, contractors should report its name and version to the client and carefully review the tool’s terms of service – in particular the provisions on ownership of intellectual property in generated content and any related rights restrictions – making sure they hold a full commercial licence from the tool’s owner.
Prompts and reference materials are where AI creation begins, as well as the leading source of data leakage and third-party infringement. To cut that risk, clients must include necessary anti-feeding clauses in their contracts.
These clauses include but are not limited to:
Given the nature of the technology, AI-generated content can pose unpredictable infringement risks. In assessing whether a client is at fault, administrative and judicial authorities focus on whether it has discharged the duty of care of an ordinary commercial party.
A client is the rights holder, user, final publisher and beneficiary of the generated content. It therefore cannot rely blindly on a contractor’s undertaking to bear infringement liability.
To demonstrate reasonable diligence, it must establish a content review mechanism, apply the “access + substantial similarity” test widely adopted in judicial practice, and review the contractor’s deliverables independently.
Chinese law and judicial practice require protected works to result from human intellectual labour. Where an AI tool produces raw output unilaterally and no human reworking follows, its originality may not be legally recognised.
A contract should therefore contain a substantive human processing clause, barring the contractor from delivering AI output that has not been reworked by a human. Terms of service for many mainstream AI tools place varying limits on the ownership and licensing of generated content. Platforms may require users to grant licences for model training or promotion, or curb exclusive ownership of the output.
To guard against platform rules leaving parts of the contractor’s deliverables without full exclusive ownership, contracts should require the contractor to disclose the platform’s intellectual property terms, and to ensure that rights in the work vest in the client.
The randomness of AI and the opacity of its generation process mean that after-the-event proof often faces a high technical evidentiary bar. Contractors should therefore be required to keep full technical records of their use of AI tools including basic logs covering the specific version used, the exact time of operation, and the identity of the operator.
The trail from input to output is just as important. Contractors should be required to record a summary of the prompts used and retain the full logical progression of those prompts through repeated fine-tuning and adjustment, together with the corresponding original outputs at each stage.
They should also keep direct evidence of human involvement, including prompt design and selection, parameter fine-tuning, manual post-editing, retouching and editing.
Contractors should also keep proper AI-labelling compliance records and, on delivery, clearly identify every part of the work generated with AI tools.
In practice, compliance disputes over intellectual property infringement and data breaches often remain latent for long periods.
To guard against the loss of key evidence when a collaboration ends or an AI platform automatically wipes historical data, contractors should preserve all material and evidence for at least three years after the collaboration, and clients should have a dedicated audit right over the contractor’s AI compliance.
Quan Wei is a partner and He Yirong is an associate at Jingtian & Gongcheng
Jingtian & Gongcheng
34/F, Tower 3, China Central Place
77 Jianguo Road, Beijing 100025, China
Tel: +86 10 5809 1026
Fax: +86 10 5809 1100
E-mail: quan.wei@jingtian.com
he.yirong@jingtian.com
www.jingtian.com
Following months of market research, we have selected these outstanding in-house counsel and their teams
CRS 2.0 and offshore trust rules tighten tax oversight as senior partners map out responses
Back in the black: 128 law firms’ revenue data
When rules conflict and resources are finite, compliance is no longer about getting it right, but making trade-offs. Law firm managers map out where to yield and where to hold for GCs
Follow us on LinkedIn
