DHS’s Responsible Use of Generative AI Tools

By Michael Boyce, Director of the DHS AI Corps

I am pleased to announce the launch of DHSChat, a new Artificial Intelligence (AI) powered chatbot designed exclusively for internal use within the Department of Homeland Security. DHS Chat is a significant step forward in leveraging secure, cutting-edge technology to enhance productivity and support our critical missions. DHSChat is now available to all the roughly 19,000 employees at DHS Headquarters and select pilot users across ten operating agencies.

Last year, we began allowing employees to use commercial generative AI tools like ChatGPT and Claude on publicly available information only; thousands of employees have now received training and use these tools to support their work. DHSChat uses the same AI technology but was built internally and operates within a secure environment. By using DHSChat, employees are able to perform routine work more efficiently. This includes summarizing complex documents and reports, generating computer code, and streamlining repetitive tasks like data entry. With this new tool, thousands of employees will be able to leverage generative AI capabilities safely and securely using non-public data. In the future, we hope to create a secure internal knowledge hub, which staff can query for information about DHS policies, data, and other internal information.

By collaborating with cloud, cybersecurity, privacy, civil rights, and civil liberties experts across the Department, DHS developed guardrails for DHSChat to ensure that it is effective, safe, secure and responsible.

While a true team effort from across the Department, DHSChat is one of many achievements from my team, the DHS AI Corps. The AI Corps hiring initiative is an effort to recruit fifty seasoned AI experts and technologists into the Department to support AI development and deployment across the DHS mission. As of mid-December thirty-nine of the fifty have started work at DHS, and eight have confirmed start dates. DHSChat is just one of twenty AI-related projects the AI Corps has undertaken since its creation this Spring.

How We Got Here

DHSChat builds on the Department's year-long effort to empower employees to take advantage of commercially available generative AI (GenAI) tools, including pilots within the Federal Emergency Management Agency (FEMA), United States Customs and Immigration Services (USCIS), and Homeland Security Investigations (HSI).

Earlier this year, we released the DHS AI Roadmap that outlined our AI initiatives and the technology's potential across the homeland security enterprise. It was the most detailed AI plan put forward by a federal agency, directing our efforts to fully realize AI's potential to protect the American people and our homeland while steadfastly protecting privacy, civil rights, and civil liberties. The Roadmap made clear that DHS must enable and encourage personnel to responsibly use commercial GenAI products to prepare our workforce for the future.

Initial steps included performing a privacy impact assessment, establishing rules of behavior, and developing an employee training. For several months we held weekly training sessions for personnel to learn how to leverage GenAI responsibly. DHS also made these resources publicly available so others could benefit from them.

As users began to leverage the approved products, we started to hear about innovative use cases that added value across DHS. Here are several of the valuable ways DHS personnel have already been using commercial GenAI tools:

• Enhancing Grant Program Communications: DHS teams use GenAI to simplify and clarify email and written communications, enhancing efficiency and increasing work capacity by 25-50%.
• Simplifying Cyber Terminology for Personnel: Cyber experts use GenAI to simplify and summarize complex cybersecurity terminology for non-cyber personnel, which enhances communication, saves time, and enables informed decision-making through clearer, accessible briefings.
• Designing Training Programs: DHS employees generated an initial outline for AI law and policy training for attorneys, providing a starting point for training designers that facilitated efficient collaboration, saved time, and enhanced the quality of the final training materials.
• Creating Study Aids: DHS staff used GenAI to create a mnemonic system for studying the National Institute of Standards & Technology (NIST) Risk Management Framework, transforming complex steps and roles into engaging stories and images, thereby enhancing retention, engagement, and successful certification outcomes.

Throughout the development of DHSChat and other generative AI tools, some of the valuable insights we have gathered include:

• Meet users where they are. Personnel may be interested in GenAI, but that does not mean they have used it before. For them to get the most out of GenAI, we must offer training that does not require users to have a strong technical foundation and gives users opportunities to experiment and ask questions.
• Talk about expected AND unexpected results. Users need to feel comfortable interacting with the tools when they produce unexpected results, including results that are wrong or that do not answer the user's prompt. Demonstrating this experience live and explaining possible reasons for the results helps empower users when they encounter similar results on their own.
• Give users access to tools. Providing access to approved commercial tools is a great way to prepare our organization for the future.

Our journey has just begun, and we look forward to continuing to innovate and lead in this space.

To learn more about the Department's Commercial GenAI Resources, visit the Commercial GenAI Resources page.

To learn more about the ways DHS is safely and responsibly leveraging AI to protect the homeland, visit the Artificial Intelligence at DHS webpage.