The U.S. Food and Drug Administration (FDA) stands at a critical technological juncture. After a period of aggressive, centralized digital transformation fueled by the rapid adoption of large language models (LLMs) and agentic AI, the agency is facing a sudden leadership vacuum. The departure of former Commissioner Marty Makary—the primary architect of the FDA’s high-speed AI integration strategy—has left stakeholders, industry insiders, and agency staff questioning whether the FDA will maintain its momentum or revert to a fragmented, cautious, and siloed approach to technology.
While acting leadership insists that artificial intelligence remains a cornerstone of the agency’s future, the sudden exodus of key architects, including Chief AI Officer Jeremy Walsh and acting Chief Information Officer Sridhar Mantha, has cast a long shadow over the agency’s internal roadmap.
Main Facts: A Paradigm Shift in Regulatory Review
At the heart of the FDA’s modernization efforts was the launch of "Elsa," a sophisticated, agency-wide AI tool designed to handle the heavy lifting of regulatory review. Unlike generic consumer-grade AI, Elsa was built as a bespoke, retrieval-augmented generation (RAG) system. By confining the model to a curated, trusted database specific to individual FDA centers, the agency sought to mitigate the risk of "hallucinations" while accelerating the evaluation of complex drug submissions.
The agency’s ambition was clear: to move beyond basic administrative automation into the realm of "agentic AI"—systems capable of performing complex tasks, from summarizing decades of regulatory history to parsing massive datasets in premarket reviews. This was not merely an upgrade; it was a fundamental shift in how the FDA handles its workload. However, with the departure of the leadership team that championed these initiatives, the governance structure surrounding these tools has become opaque. Experts now worry that the lack of a centralized "AI Czar" will lead to a regression, where individual divisions—such as the Center for Drug Evaluation and Research (CDER) or the Center for Biologics Evaluation and Research (CBER)—begin building their own disparate, incompatible digital infrastructures.
Chronology: From Rapid Acceleration to Institutional Uncertainty
The trajectory of the FDA’s AI program has been marked by extreme volatility over the past two years:
- 2024 (The Great Expansion): Under Commissioner Makary, the FDA reported a 148% surge in AI use cases, according to the Bipartisan Policy Center. The agency pivoted from exploratory pilot programs to enterprise-level integration.
- December 2025: The FDA formally announced the expansion of agentic AI capabilities for premarket reviews and inspections, signaling an intent to weave automation into the core of the approval process.
- Early 2026: A wave of high-level resignations hits the FDA, including Commissioner Makary, Chief AI Officer Jeremy Walsh, and acting CIO Sridhar Mantha.
- May 2026: Acting Commissioner Kyle Diamantas assumes the helm. He moves to distance the agency from the "shoot-from-the-hip" policymaking style of his predecessor, reaffirming that informal pronouncements made via journal articles or press conferences do not constitute official regulatory policy.
- Present Day: The agency enters a period of administrative transition, where the focus has shifted from rapid deployment to ensuring policy compliance and institutional stability.
Supporting Data: The Scale of Integration
The Bipartisan Policy Center’s tracking of federal health agencies highlights a dramatic, nearly unprecedented adoption curve. Between 2024 and 2025, the sheer volume of documented AI use cases at the FDA grew by 148%. This data suggests that AI is no longer a peripheral experiment but a critical component of the FDA’s operational backbone.
The efficacy of these tools, particularly the evolution of "CDER GPT" into the broader Elsa platform, has been documented as a success in terms of efficiency. By providing reviewers with instant access to summarized industry comments and historical regulatory submissions, the agency has significantly reduced the time spent on document retrieval and manual synthesis. Despite this, the lack of public-facing documentation regarding these algorithms remains a point of contention. There is currently no unified, public-facing ledger that details exactly how these AI models are being used to "assist" in the final stages of regulatory review.
Official Responses and the Return to Norms
The current administration, led by Acting Commissioner Kyle Diamantas, is signaling a clear pivot toward traditional governance. In recent briefings, the agency has emphasized that while AI remains a priority, it must operate within the established framework of the FDA’s formal guidance processes.
This is a direct response to the criticism leveled at the previous administration, which often utilized non-traditional channels to announce shifts in AI strategy. Industry leaders, while appreciative of the speed that characterized the Makary era, are now navigating a more conservative environment. The agency has explicitly disavowed past informal statements, clarifying that the FDA’s regulatory requirements for drug sponsors remain unchanged.
Tala Fakhouri, Chief AI and Regulatory Strategy Officer at Parexel and a former FDA AI policy official, notes that this return to "regular order" is a double-edged sword. "While traditional rulemaking provides the stability that pharma companies crave, it is inherently slow," she explains. "In the age of generative AI, where technological capabilities double in performance every few months, a one-year guidance development cycle is effectively an eternity."
Implications: The Transparency and Policy Gap
The implications of this leadership transition are profound, particularly regarding the relationship between the regulator and the regulated industry.
The Transparency Deficit
There is a growing demand for the FDA to be more transparent about how its internal AI tools interact with sponsor submissions. Industry stakeholders are concerned that if they do not know how an AI "assistant" interprets their data, they cannot optimize their submissions to ensure the most accurate review. Transparency would allow sponsors to tag data and provide information in ways that align with the FDA’s AI-driven assessment protocols, ultimately leading to faster and more predictable outcomes.
The Validation Challenge
A major lingering question is the validation of AI tools used in clinical trials. As pharmaceutical companies increasingly adopt AI for predictive analytics, patient stratification, and endpoint monitoring, the FDA must establish clear rules for how these tools should be validated. Without clear guidance, companies face the risk of investing millions in AI technologies that may be rejected during the premarket review phase due to a lack of regulatory consensus.
The Need for Agile Governance
The fundamental challenge facing the FDA is how to maintain the "human-in-the-loop" necessity for final decision-making while adopting the agility required by modern software cycles. The current consensus among experts is that the agency cannot afford to retreat into silos. Instead, it requires a new, hybrid model of governance—one that is as agile as the technology it regulates, yet as rigorous as the public health mandate it serves.
"We should all be happy that staff are using these tools to augment their work," says Fakhouri. "The risk is not that the technology is being used; the risk is that the lack of a clear, centralized vision will create a fragmented regulatory landscape where different divisions move at different speeds, creating uncertainty for everyone involved."
As the FDA navigates this period of institutional soul-searching, the path forward appears to rely on a delicate balance: the agency must move quickly enough to remain relevant in a rapidly digitizing global healthcare economy, while resisting the temptation to bypass the rigorous, peer-reviewed, and public-facing processes that have defined its authority for nearly a century. Whether the new leadership can successfully institutionalize the AI progress made during the Makary era without the same volatility remains the defining question for the agency’s next chapter.
