By admin 
The integration and evolution of artificial intelligence within the U.S. Food and Drug Administration (FDA) appear to be at a critical juncture, following a series of high-profile departures, including former commissioner Marty Makary, who was a driving force behind a more centralized AI strategy. While acting commissioner Kyle Diamantas affirms AI remains a top priority, the recent leadership vacuum has cast a shadow of uncertainty over the agency’s internal AI development and governance structure, potentially signaling a return to a more fragmented, division-specific approach.
Makary’s Vision: A Centralized AI Hub
During his tenure, former commissioner Marty Makary championed a vision of a unified, agency-wide AI infrastructure designed to revolutionize the FDA’s operational efficiency. His proactive stance led to the rollout of "Elsa," a large language model (LLM) developed internally, aimed at accelerating critical processes such as regulatory reviews and evaluations. Elsa was conceived to handle support tasks, including drafting summaries, generating reports, and retrieving historical data, thereby freeing up human reviewers for more complex analytical work. The agency further underscored its commitment in December with plans to expand the use of agentic AI—autonomous AI systems capable of executing specific tasks—to bolster premarket reviews, inspections, and various administrative functions.
This push for centralization under Makary was not an isolated initiative but part of a broader, government-wide mandate to leverage advanced technology for improved public service. The impact of this directive was palpable: between 2024 and 2025, the number of AI use-cases reported by the FDA witnessed a remarkable 148% surge, according to a study by the Bipartisan Policy Center. This rapid expansion reflected an organizational pivot towards harnessing AI to manage the increasing volume and complexity of scientific data and regulatory submissions inherent in modern drug development and oversight. Makary’s strategy aimed to consolidate these efforts, ensuring consistency, maximizing resource allocation, and fostering a shared understanding of AI’s potential across the FDA’s diverse centers.
The Departure Cascade: A Leadership Void
The ambitious trajectory set by Makary now faces headwinds. His departure, alongside that of other key AI architects, has created a significant leadership void. Jeremy Walsh, the FDA’s chief artificial intelligence officer, and Sridhar Mantha, the acting chief information officer, have also stepped down. These simultaneous exits of top AI leadership have prompted serious questions about the continuity and future direction of the agency’s AI implementation strategy.
Tala Fakhouri, a former FDA AI policy official and current chief artificial intelligence and regulatory strategy officer at Parexel, articulated these concerns, stating, "It’s unclear now what the leadership and governance structure are around FDA-wide efforts." Fakhouri observed that recent FDA officials speaking at conferences about AI have primarily represented individual divisions, rather than presenting a unified agency-wide vision. This shift could presage a regression to the more siloed, department-specific AI initiatives that characterized the agency’s approach prior to Makary’s centralization efforts. Such a fragmentation, while allowing for tailored solutions, could impede the development of overarching AI policies, hinder data sharing across divisions, and slow down the pace of broader AI adoption and innovation.
Internal AI’s Evolving Role and Current Functionality
Despite the leadership changes, the FDA’s internal use of AI to streamline staff workloads has seen steady evolution. The "Elsa" program, initially rooted in a CDER-developed program known as CDER GPT, exemplifies this progression. The agency enhanced this original concept by incorporating a retrieval-augmented generation (RAG) system. This sophisticated architecture is designed to mitigate "AI hallucinations"—instances where large language models generate plausible but incorrect or nonsensical information—by confining the LLM’s knowledge base to a meticulously curated database of trusted, agency-specific information. This customized approach allows the system to provide highly relevant and accurate information, tailored for the unique needs of individual centers within the FDA.
Elsa empowers staff members to efficiently access critical information pertinent to their specific roles. For instance, it can rapidly summarize industry comments on a particular regulatory proposal, providing a concise overview that would otherwise require extensive manual review. Similarly, the Office of New Drugs can leverage Elsa to quickly compile a comprehensive history of regulatory submissions for a specific product, a task that historically could span years of documentation. These applications highlight AI’s capacity to significantly reduce the burden of repetitive, time-consuming tasks. Fakhouri emphasizes that while Elsa is instrumental in augmenting staff work, it is not currently employed for final decision-making processes, maintaining the crucial human oversight in regulatory judgments. "Staff can use the tools to augment the work that they’re doing. We should all be happy about that," she noted, underscoring the assistant role of AI.
External AI Policy: Stability Amidst Internal Flux
Crucially, the potential internal shifts in the FDA’s AI strategy primarily concern the agency’s own operational use of the technology. These changes are distinct from, and do not currently impact, the FDA’s policies governing how pharmaceutical companies and other regulated entities apply AI in their own work, such as in drug discovery, clinical trials, or manufacturing. Fakhouri reassured that there hasn’t been a change in the FDA’s policies related to sponsor use of AI, and she doesn’t anticipate one. This stability is a significant positive for the industry, providing a predictable regulatory environment for companies investing heavily in AI-driven innovation.
The FDA has been actively engaged in developing guidance for industry on the responsible use of AI in medical product development. This includes, for example, considerations for AI/Machine Learning-based medical devices and the use of AI in clinical trial design and monitoring. The consistency in these external policies allows pharmaceutical companies to continue integrating AI into their workflows with a clear understanding of regulatory expectations, fostering innovation without the added burden of shifting compliance goalposts.
The Path Forward: Transparency and Agile Regulation
Despite the internal uncertainties, a consensus among experts points to the critical need for increased transparency and more agile policymaking from the FDA regarding AI. Fakhouri advocates for greater clarity on how the FDA itself utilizes AI in its processes. "If the regulators are using AI in certain ways to augment reviewer work or to become an assistant to a reviewer, I think it’s good practice for industry to know what these uses look like," she asserted. This transparency is not merely about accountability; it could foster greater collaboration and efficiency.
Imagine a scenario where a sponsor or a Contract Research Organization (CRO) preparing a submission understands how the FDA’s AI assistants process information. This knowledge could enable them to structure their submissions with optimal data labels and information formats, making the review process smoother and faster for both human reviewers and AI tools. Fakhouri expressed optimism that the agency will move towards greater transparency over time, contingent on leadership prioritization. "But it would require someone in a leadership position at the agency to become aware of that and want to actually make it happen," she added.
Beyond internal transparency, Fakhouri also calls for streamlining AI-related rulemaking that directly impacts the industry, such as guidelines for validating AI tools used in clinical trials. Under former commissioner Makary, some policy changes were announced through unconventional channels like journal articles or press conferences, diverging from the traditional FDA guidance process. However, the agency appears to be reverting to its established norms under acting commissioner Diamantas, who recently confirmed that informal statements by the former commissioner do not represent official policy. Fakhouri noted, "They will go through the regular guidance and policy development processes."
While a return to traditional rulemaking offers predictability and insulates pharma companies from regulatory uncertainty, it presents its own set of challenges in the rapidly evolving landscape of AI. The inherent slowness of the traditional guidance development process—which can take a year or more—contrasts sharply with the blistering pace of AI innovation. "A year in the age of AI is very slow," Fakhouri observed. This disparity creates a dilemma: how can the FDA provide stable, clear guidance without stifling the rapid adoption of new AI platforms by the industry?
Balancing Structure and Flexibility: A Key Policy Challenge
The FDA’s central policy challenge in the coming years will be to strike a delicate balance between providing robust regulatory structure and maintaining the flexibility necessary to accommodate the rapid advancements in AI. This necessitates a re-evaluation of current policy development mechanisms. Fakhouri suggested the need for "more agile, flexible policy development" and "perhaps more frequent communication between industry and the regulators." This could involve new engagement models, such as ongoing public workshops, pilot programs, or even "living guidance" documents that can be updated more frequently than traditional regulations.
The broader context of government-wide AI adoption further complicates this. The Biden administration has pushed federal agencies to embrace AI, issuing executive orders that mandate responsible AI use and inventorying AI systems. The FDA, as a critical public health agency, is under pressure to not only adopt AI internally but also to regulate its use in a complex, high-stakes industry. The internal leadership shuffle, therefore, reverberates beyond mere operational efficiency; it impacts the agency’s capacity to lead or keep pace with global regulatory bodies in defining the future of AI in healthcare.
The future direction of the FDA’s internal AI program remains a subject of considerable interest and speculation. While the commitment to AI from the acting commissioner is clear, the practical execution and governance structure will be pivotal. The agency must navigate the tension between maintaining regulatory rigor and fostering innovation, ensuring that AI serves as a powerful tool for public health without compromising safety or trust. How the FDA resolves its internal leadership questions and adapts its policymaking to the speed of AI will ultimately define its effectiveness in the age of intelligent machines.
