AI News Bureau
Written by: CDO Magazine Bureau
Updated 3:51 PM UTC, Tue January 21, 2025
Paula Osborn, Deputy Chief Data and AI Officer, at the U.S. Department of State (DOS), speaks with Amy Jones, U.S. Public Sector AI Lead at EY, in a video interview about approaching change management for successful AI adoption, using generative AI (GenAI) on sensitive data, ensuring the security of sensitive data, and addressing the growing system demand and the challenges that come with it.
The U.S. Department of State is an executive department of the U.S. federal government responsible for the country’s foreign policy and relations.
When asked about change management for successful AI adoption, Osborn states that change management is critical. She highlights that many people focus on technology and acquiring it, moving on to the next thing while neglecting change management.
However, for Osborn, that is the most interesting part because it’s challenging, and reflecting on the past failures, the Department of State approaches change management comprehensively — top-down, bottom-up, and side-to-side.
Eluding to a GenAI discussion between the department’s CDAIO and the State Secretary, Osborn states that for the first time in decades, a State Secretary has discussed a tech management issue rather than a foreign affairs topic. This top-down messaging is reinforced by the department’s deputy secretaries, who are also engaged, she adds.
Next, Osborn appreciates the leadership’s involvement and states that the department ensures that their communications emphasize their commitment to leveraging these technologies. From the bottom up, the department has been identifying data champions.
For instance, she says, the State Department has an AI Innovators Teams channel with thousands of employees. It is a safe space for staff-level collaboration on projects, questions, and successes.
Referring to the Center of Analytics, Osborn states that it has a robust analytics team including data scientists, supporting the development of data analytics and AI projects aligned with the administrative priorities.
The department wants to keep people excited by developing the tools that they want while also being open to integrating AI and data into their work, but it is not an easy feat. The reason, she says, is that the state department mostly attracts humanities majors who are not necessarily technically inclined but are now getting more on board.
Shedding light on being the first federal agency to use GenAI on sensitive data, Osborn states that when ChatGPT was first released, there was a massive debate on how to leverage it. To approach it strategically, the department integrated it into the enterprise data strategy through data campaigns.
Instead of focusing narrowly on areas like data management, governance, culture, or analytics, the department conducts two broad campaigns every six months. These campaigns align with administration priorities, such as climate or multilateralism, including work with the United Nations.
Around the time ChatGPT emerged, the department had a campaign focused on public diplomacy, which was particularly well-suited for ChatGPT’s capabilities. For example, media monitoring and summarization were a natural fit for the tool.
Local staff often spend entire days sifting through news, identifying relevant content for diplomats, translating it into English, and distributing it. With ChatGPT, this process can be completed in seconds. Tasks like translation, media summarization, and similar activities also saw potential improvements.
This campaign highlighted the tool’s practical applications, and while the department stated clearly that only unclassified data could be input into ChatGPT, users expressed interest in applying it to sensitive data.
Post that, the leadership at the Center for Analytics acted swiftly and procured a secure platform faster than anticipated. This platform, known internally as State Chat, runs on the department’s sensitive systems, and its introduction has generated extraordinary enthusiasm.
To ensure the safety and security of putting sensitive data in ChatGPT, Osborn emphasizes a simple guideline of not putting anything into ChatGPT that one would not put into Google Search.
Apart from that, the Department of State also relies on internal policies that define and distinguish between unclassified and sensitive but unclassified information, detailing the various levels of sensitivity.
Now, ChatGPT is integrated into the sensitive system, but not in the higher-level classified environment, and users are familiar with the appropriate information that goes into it, says Osborn.
Nevertheless, the department continues to direct users to the internal policy to ensure only sensitive but unclassified information is entered. Since the platform operates internally, all inputs and outputs are internal, monitored, and controlled, she notes.
Thereafter, Osborn states that the demand for this system is growing, and the aim is to have it operational on the classified systems by the end of the fiscal year.
The primary concern, as always, is security, says Osborn. While the State Department is already working extremely closely with diplomatic technology and diplomatic security teams, this effort involves an additional level of scrutiny and coordination, she concludes.
CDO Magazine appreciates Paula Osborn for sharing her insights with our global community.