Using Generative AI (ChatGPT) for Corporate Compliance

Corporate regulatory compliance, especially in the field of international trade controls, poses significant challenges for companies. The constantly changing regulations and the potential for severe penalties for non-compliance can put companies at significant risk. However, generative AI like ChatGPT can help address these challenges by automating compliance-related tasks and generating compliance-compliant communications.

Use Case 1: Automated Export Control Classification:

Generative AI can be used to automatically classify products and goods for export control compliance. This process involves identifying the nature of the product, determining the relevant regulations, and classifying the product accordingly. Automated classification can significantly increase efficiency and reduce the risk of non-compliance. With the ability to process large amounts of data quickly, generative AI can classify products faster and more accurately than human counterparts.

Use Case 2: Generating Compliance-compliant Documentation:

Generative AI can also be used to generate compliance-compliant documentation such as export licenses, shipment declarations and end-user statements. This can help companies ensure that all documentation is compliant with regulations and company policies. Generative AI can be trained on a company’s specific compliance requirements, allowing for the generation of customized documents that meet all necessary regulations.

Use Case 3: Compliance Monitoring and Reporting:

Generative AI can be used to monitor and report on compliance-related data, such as trade transactions and end-use activities. This allows companies to identify and address compliance issues in a timely manner. By analyzing large amounts of data, generative AI can detect patterns and anomalies that may indicate non-compliance, and alert the relevant parties. This can help companies take proactive measures to address compliance issues before they become a problem.

While generative AI can be a powerful tool for corporate regulatory compliance, there are some limitations to be considered. One of the main challenges is the need for high-quality data. Without accurate and up-to-date data, generative AI may not be able to accurately classify products or generate compliant documents. Additionally, the complexity of regulations can make it difficult for generative AI to keep up with the constantly changing compliance requirements.

Generative AI can be a powerful tool for corporate regulatory compliance, especially in the field of international trade controls. It can automate compliance-related tasks, generate compliance-compliant communications, and help companies identify and address compliance issues in a timely manner. However, it is important to keep in mind that generative AI is not a replacement for human oversight, but rather an additional tool to be used in conjunction with existing compliance programs. Companies can get the best results by pairing generative AI with a robust compliance program that includes human oversight and review, as well as regular monitoring and reporting. This way, companies can benefit from the speed and accuracy of generative AI while still maintaining the necessary level of human oversight and expertise to ensure compliance.

I was starting to write a brief article on how to use ChatGPT for corporate compliance but thought, why not ask ChatGPT to write it for me? The above article was written entirely, word-for-word, by ChatGPT with prompts and some formatting from yours truly. Does ChatGPT have it right? Yes and no:

·      Classification – ChaptGPT or any other general, pre-trained generative AI cannot (yet) be used for product classification. As detailed in my previous post, I tried and failed. The issue is, as ChatGPT points out, one of data and dynamics.  First, you need a really large corpus of high-quality classification data to effectively train an AI model and I’m not sure one for export classification data exists in the public domain (someone please point me to one if it does exist). Even in large industrial companies that might have such a corpus, the data is usually poor-quality and often unreliable. Even generative AI is not immune to the old maxim “garbage in garbage out”. Second, export/import classification lists (e.g., USML and CCL) change over time, sometimes slowly and sometimes very quickly. If such changes are not accounted for (e.g., with a built-in recency bias), automated classifications could be a recipe for disaster. Sadly, I doubt “ChatGPT told me this product was EAR99” will be a viable mitigating factor in disclosures to DDTC or BIS.

·      Documentation and Compliance Monitoring – This should be pretty feasible with current AI technologies. The key, as ChatGPT points out, is availability of data. Currently, large enterprises run on a hodgepodge of information and compliance management systems. In order for AI to be truly useful for documentation and compliance monitoring, companies will have to break through vendor data storage barriers. Further, ChatGPT and similar AI tools are built on language models. I imagine structured data without contextual meaning provided may be difficult for such models to make use of. I’ll have to do more research on this point.

In fact, there are many other corporate compliance scenarios that I can think of that current generative AI, with some data training and prompt engineering, can readily accomplish:

·      Training and first-line compliance help desk – “Do I need an export license to send our widget to China, for use by acme co.?”

·      Entity Diligence – “I’m planning to conduct business with acme co. in Shanghai. Are they affiliated with the Chinese military?”

·      Transaction level error checking – AI automatically provides first-level error checking on transactions to ensure they meet requirements of authorization conditions, regulations, laws, etc.

·   Internal Investigations – Natural language prompting to an AI system that can then gather relevant data and communications related to a topic(s) under investigation. Of course, there may be some privacy concerns here but nothing practitioners in the space haven’t seen before.

Much work will have to be done to make the above a reality, starting with a data capture, ETL, and model training/updating framework. Perhaps work done for existing BI and corporate search tools can be reused? Also, given the strict liability that is imposed in many of these areas, sadly an “all clear” from AI may likely not be enough until the technology is widely tested, independently validated (AI audits anyone?), and adopted. Once that happens though, an appropriately trained AI model could be useful in all aspects of a business, from finance to procurement and logistics, to personnel management; truly, one’s imagination may be the only limit.

ChatGPT and similar models/tools are a truly momentous breakthrough (ChatGPT can already pass portions of a bar exam – see work by Michael Bommarito and Daniel Katz) and Microsoft and Google are already making big bets on (or in response to) ChatGPT. It’s only a matter of time that this technology is integrated into Big Tech product ecosystems and business information system vendors and integrators won’t be far behind.