On 30 October 2023 the US President signed Executive Order 14110 on the safe, secure and trustworthy development of artificial intelligence. The order is long, covering frontier model safety testing, labelling of AI-generated content, privacy, consumer protection, labour effects and federal use of AI.
For US companies it is directly applicable. For Chinese AI companies, its influence is mainly indirect, through supply chains, cross-border services and customer requirements.
Frontier models and safety testing
The order requires developers of large frontier models to report to the federal government before training and to conduct safety testing before release, including red-teaming and capability evaluations. Triggers are generally based on compute scale.
Compliance costs for large model development rise, with possible effects on open and closed strategies. Chinese companies using US compute, cloud services or foundation models may be asked by customers and suppliers for safety assessment documentation.
Labelling generated content
The order calls for standards and tools to identify AI-generated content, including watermarking, metadata and content credentials. The direction parallels China’s deep synthesis labelling and generative AI content labelling requirements.
For product teams, labelling should be built in rather than bolted on. Images, video, audio and text require different methods, and cross-border platforms must consider labelling formats and legibility in each jurisdiction.
Privacy and consumer protection
The order requires assessment of AI privacy impacts, promotion of privacy-enhancing technologies and stronger consumer protection. Scrutiny increases where personal information is used for training, automated decisions and profiling.
For Chinese companies, providing AI services in the US requires attention to state privacy laws, federal consumer protection and sector rules. A product designed only around Chinese PIPL may not satisfy US notice, choice and deletion requirements.
Federal use of AI
The order sets requirements for federal procurement and use, including risk assessment, human oversight and transparency. Procurement transmits these requirements to suppliers through contract terms, creating de facto standards.
Companies supplying AI to US public bodies should examine security assessment, data use, audit rights and incident reporting clauses, which are often stricter than ordinary commercial contracts.
Three effects on Chinese companies
First, supply chain transmission. Companies using US chips, cloud services or models may be asked to demonstrate that they are not used for restricted purposes. Second, market access. AI products entering the US must assess sector regulation and state law. Third, contract terms. Cross-border contracts may include safety assessment, labelling, localisation and audit clauses.
What can be prepared
First, map AI functions and model sources, identifying frontier models or controlled compute. Second, build content labelling capability and documentation. Third, assess privacy and consumer protection requirements for US business. Fourth, review data use, audit and incident clauses in cloud and model supplier contracts. Fifth, integrate AI safety assessment into product launch.
For AI product launch and documentation, see our AI product launch documents service; for cross-border AI and multi-jurisdiction compliance, see multi-jurisdiction; for compute and export controls, see export controls.
This article is based on US Executive Order 14110 of 30 October 2023. It is for reference only and does not constitute legal advice.