Global AI Large Model Compliance Regulatory Guidelines: Core Regulatory Framework and Implementation Points

In today's rapid development of artificial intelligence, governments have introduced regulations specifically for AI, especially generative AI. These regulations form the core framework of AI compliance, directly stipulating the full lifecycle requirements of AI systems, from development to deployment to operation.

1. Regulations specifically for AI/generative AI (most directly related)

China's Interim Measures for the Administration of Generative AI Services

⭐⭐⭐⭐⭐ the most relevant: specifically regulating generative AI, Including the core requirements of the whole process of training data, content generation, and service provision

: algorithm filing, security assessment, content filtering, and identification obligations

Official website: http://www.cac.gov.cn/2023-07/13/c_1690898327029107.htm

the most relevant ⭐⭐⭐⭐⭐ the EU's

AI Act

: the world's first comprehensive AI regulatory law, There are specific core requirements for the basic model

: risk assessment, transparency obligations, technical documentation, manual supervisionOfficial

website: https://digital-strategy.ec.europa.eu/en/policies/regulatory-framework-ai

2. Algorithm governance related (directly affecting model deployment)

China

"Provisions on Algorithm Recommendation of Internet Information Services" ⭐⭐⭐⭐

Official website for

algorithm filing, algorithm transparency, and protection of user rights

and interests: http://www.cac.gov.cn/2022-01/04/c_1642894606364259.htm

"Provisions on Deep Synthesis of Internet Information Services" ⭐⭐⭐⭐

Official website for

deep synthesis content identification, detection technology, and prevention of illegal content

: http://www.cac.gov.cn/2022-12/11/c_1672221949354811.htm

3. Data compliance (related to training data)

Global

GDPR (EU) ⭐⭐⭐⭐⭐

If training data involves personal information, it requires legal basic

data subject rights (right to delete, right to access, etc.).

Official website: https://gdpr.eu/

China's

Personal Information Protection Law ⭐⭐⭐⭐⭐

the legality of training data collection and use

Special protection of sensitive personal information

official website: http://www.npc.gov.cn/npc/c30834/202108/a8c4e3672c74491a80b53a172bb753fe.shtml

Data Security Law Important ⭐⭐⭐⭐

Data Processing Rules

Data Export Security Assessment

Official Website: http://www.npc.gov.cn/npc/c30834/202106/7c9af12f51334a73b56d7938f99a788a.shtml

4. Content security (related to output content)

China's

Cybersecurity Law Monitoring ⭐⭐⭐⭐

and Disposal of Illegal Content Network

Security Level Protection

Official Website: http://www.cac.gov.cn/2016-11/07/c_1119867116.htm

"Provisions on the Ecological Governance of Online Information Content" Official ⭐⭐⭐

website of the

review requirements

for

the content of bad information governance

: http://www.cac.gov.cn/2019-12/20/c_1578375159509309.htm

5. Intellectual property rights (related to training data and output)

Copyright Law (available in all countries). ⭐⭐⭐⭐

Copyright compliance for training data

Copyright attribution of generated content

6. Industry-specific norms

Financial sector

China: Fintech Development Plan, relevant regulations of the People's Bank of China

European Union: Digital Operational Resilience Act (DORA)

Medical field

United States: FDA's AI/ML medical device guidelines

China: Summary of core compliance points of AI-assisted diagnosis related regulations

in the Medical Device Classification

Catalogue Algorithm Filing (China): To provide generative AI services, algorithm filing security assessment must be

completed: security assessment before launch, Including content security, data security

training, data compliance: ensuring that data sources are legal and does not infringe on intellectual property

rights Transparency requirements: informing users to use AI-generated content, and technical documentation disclosure

content filtering: preventing the generation of illegal and harmful content

Personal information protection: privacy protection during training and reasoning

Manual supervision: Establish a manual review mechanism

Risk classification management: Conduct risk assessment and corresponding management

suggestions according to application

scenariosFocus on:

If operating in China: Prioritize the Interim Measures for the Management of Generative AI Services If

operating

in the EU: Prioritize the AI Act and GDPR If

processing personal data: Data protection laws in each country need to be complied with

If it involves a specific industry: also need to pay attention to industry regulatory requirements