The U.S. Food and Drug Administration (FDA) is navigating significant internal disruptions following the hasty deployment of its artificial intelligence assistant, Elsa. Intended to revolutionize operational workflows and boost agency-wide efficiency, the system has instead encountered a range of technical problems that have sparked concern and confusion among staff.
Launched on June 2, 2025, ahead of its projected timeline, Elsa represents the FDA’s most ambitious foray into generative AI. The tool was expected to streamline document reviews, summarize adverse event reports, assist with clinical data coding, and guide regulatory decision-making. However, since its rollout, numerous issues have surfaced, particularly relating to system functionality and user accessibility.
Many employees have reported being unable to upload essential documents or access internet-dependent features within the tool. Additionally, there have been widespread complaints about the failure of Elsa to process queries reliably. These early difficulties have impaired productivity and reduced trust in the system, raising red flags about the broader implications of accelerating AI deployment without comprehensive testing.
Commissioner Marty Makary has acknowledged the tight schedule that accompanied Elsa’s launch, noting the administration’s goal to implement AI tools across all FDA departments by the end of June 2025. While Makary lauded the project for coming in ahead of schedule and under budget, critics within the agency argue that speed may have come at the cost of stability and user readiness.
FDA Chief AI Officer Jeremy Walsh remains optimistic. He described Elsa as the beginning of a transformative era for the agency. Walsh emphasized that the tool is still in its formative stage and improvements will be implemented rapidly. He also highlighted the importance of maintaining human oversight and judgment, positioning Elsa as a supportive—not substitutive—system.
Yet internal feedback has not been uniformly positive. Staff within the Center for Devices and Radiological Health, one of the first divisions to integrate Elsa, have expressed specific concerns. Users in this division reported that Elsa often fails to load key documents and provides incomplete responses to technical questions. In some cases, the interface itself became unresponsive during peak usage periods.
Elsa was built on a secure government cloud infrastructure designed to handle sensitive FDA data while adhering to strict privacy and cybersecurity requirements. While this security-focused approach is essential for safeguarding regulatory information, some have questioned whether it inadvertently hampers performance, particularly when real-time connectivity is crucial.
The rollout also comes amid a broader federal push to integrate AI technologies into public service. Across various agencies, leaders are seeking to capitalize on the efficiencies promised by generative AI. However, Elsa’s rocky start highlights the complexities of translating theoretical innovation into functional, day-to-day tools for civil service workers.
Some agency insiders argue that Elsa’s difficulties are not purely technical but stem from a lack of user-centered design and training. Employees were reportedly given limited onboarding support, with some departments lacking dedicated sessions or manuals explaining how to interact effectively with the tool. As a result, many users remain uncertain about Elsa’s capabilities and limitations, leading to underutilization or misuse.
Despite the glitches, there is a roadmap for improvement. The FDA plans to release several updates over the coming months, focusing on enhancing Elsa’s ability to handle various document formats, refining its response accuracy, and improving the overall user interface. The agency also intends to gather ongoing feedback from employees to guide these refinements.
The Elsa rollout is being closely watched by other federal entities considering AI integration. Its challenges serve as a cautionary tale about the need for deliberate, iterative deployment strategies that prioritize usability, reliability, and staff preparedness.
Moving forward, the FDA aims to strike a balance between innovation and practicality. While the agency remains committed to technological modernization, it acknowledges that successful AI integration requires not just cutting-edge algorithms, but also a workforce equipped and willing to embrace change. Elsa may yet fulfill its promise—but only if the FDA can reconcile visionary ambitions with the operational realities of a complex, high-stakes regulatory environment.
