IMF-用于合规风险分析的生成性人工智能_在税务和海关管理中的应用（英）-2025.8_63页_1mb

Summary of Generative Artificial Intelligence for Compliance Risk Analysis

This document from the International Monetary Fund (IMF) explores generative artificial intelligence (GenAI) and its applications in tax and customs compliance risk analysis. Authored by Joshua Aslett, Thomas Cantens, François Chastel, Emmanuel Crown, and Stuart Hamilton, it addresses key questions on GenAI's functionality, operational use, and future developments in the field.

Introduction and Purpose

The note raises awareness of GenAI by explaining and demonstrating its capabilities in compliance risk analysis for tax and customs administrations. It highlights GenAI's transformative potential while cautioning about risks, such as data confidentiality and ethical considerations. Deployment options range from commercial hosted services to on-premises solutions, offering flexibility for different administrative contexts.

Understanding Generative AI

GenAI is defined as AI designed to generate human-like content, such as text, images, or data, using advanced machine learning techniques like large language models (LLMs) and deep neural networks. It works through foundation models trained on vast datasets, enabling natural language interactions and tasks like research, analysis, and analytics. Key components include inputs, parameters, training data, and interaction methods.

Applications in Compliance Risk Analysis

GenAI can augment risk analysis by handling tasks like natural language processing, data interpretation, and predictive analytics. Practical use cases include:

• Assistant AI: Summarizing documents or automating basic tasks.
• Consultant AI: Providing deeper insights from deep expertise.
• Collaborator AI: Assisting in research and analysis for risk reviews.
• Autonomous Agent AI: Executing tasks independently without human oversight.

Demonstrations

The document presents experimental demonstrations using hosted GenAI services, focusing on research, analysis, and analytics. For instance:

• Research: Analyzing economic crises to identify tax risks.
• Analysis: Reviewing multinational company reports for tax compliance issues.
• Analytics: Using GenAI to predict and rank high-risk tax cases.

On-premises applications are also discussed, leveraging systems like ASYCUDA for integration, with examples of agentic integrations for data analysis, showcasing improved efficiency in handling structured data.

Suggested Guidelines for Operational Use

To ensure responsible implementation, the IMF recommends:

1. Understand When to Use GenAI: Limit use to cases where GenAI offers a clear advantage.
2. Mandate Human Accountability: Ensure humans remain responsible for decisions and explainable outcomes.
3. Anticipate Changes and Prioritize Training: Prepare for organizational shifts and invest in staff training.
4. Build and Secure a Compliance Repository: Create secure, accessible databases for GenAI use.
5. Iterate Adoption Slowly: Test GenAI cautiously with small-scale, risk-assessed use cases.

Conclusion

GenAI has the potential to revolutionize compliance risk analysis by enabling new roles like assistants and collaborators, but it must be managed carefully to address privacy, bias, and security concerns. Future developments include wider integration and democratization of AI tools. GenAI will reshape tax and customs analysis, emphasizing human oversight and ethical governance.

• Systemic Impact: May lead to new tax evasion forms and widen the digital divide.
• Recommended Actions: Focus on ethical use and continuous risk assessment.