Key Takeaways
- Regulatory Acceleration: The FDA is integrating generative AI to streamline scientific reviews, reducing some tasks from days to minutes.
- New AI Tools: The agency launched "Elsa" to assist with clinical protocols and is exploring "cderGPT" in collaboration with OpenAI.
- Efficiency Gains: AI integration aims to save hundreds of thousands of review hours annually at the Center for Drug Evaluation and Research (CDER).
- Precision Medicine Impact: Faster regulatory cycles will reduce time-to-market for complex, targeted therapies and gene treatments.
- Industry Adoption: Companies like Sano Genetics are already piloting internal AI assistants to automate clinical study design and protocol drafting.
Why now?
The FDA’s recent adoption of AI comes at a critical juncture for medical innovation. Scientific momentum has been building significantly over recent years, with breakthroughs in gene therapy, cell-based treatments, and RNA technologies. These have transformed outcomes among patients with rare and previously untreatable conditions.
At the same time, the FDA has been receiving a growing number of drug application submissions that include AI components, spanning nonclinical, clinical, postmarketing, and manufacturing phases. The agency is not only exploring AI for its own workflows but responding to a shift already underway across the industry.
Technological innovations must be matched by agility on the regulatory side. Since many of these new therapies are developed for small populations with high unmet need, time is of the essence. If the science exists to develop these treatments quickly, the regulatory framework must evolve to evaluate them just as rapidly.
At the same time, AI technologies have matured to the point where they can augment and automate complex regulatory tasks. Advances in large language models, knowledge retrieval, and document synthesis now allow agencies to restructure how safety, efficacy, and compliance are assessed across the drug development lifecycle. In this context, the question is no longer whether AI belongs in regulatory science, but how to deploy it responsibly given the growing volume and complexity of modern drug development submissions.
FDA uses AI to assist in scientific review for the first time
The FDA's scientific review process is by design rigorous and document-intensive, as it forms a critical checkpoint in drug evaluation. However, many of the required tasks are repetitive and amenable to automation. This motivated the decision to test the performance of generative AI in assisting experts to complete scientific reviews.
Initial results indicated meaningful efficiency gains : certain review tasks that previously required three days were completed in minutes, reducing time allocated to structured, repeatable steps and redirecting reviewer capacity toward higher-judgment work. FDA leadership highlighted that this would reduce non-productive tasks that were previously part of the process, therefore freeing up time for tasks requiring deep expertise and recognizing the value of their scientists’ time.
FDA releases Elsa and announces wider plans for AI
Alongside the announcement of the pilot program, the FDA aimed to scale up AI integration across the agency by the end of June. This included expanding use to other processes, improving performance, and adapting functionality based on the needs of each center.
Importantly, these internal efforts were accompanied by external-facing policy. In 2025, the FDA published a draft guidance titled 'Considerations for the Use of Artificial Intelligence to Support Regulatory Decision Making for Drug and Biological Products,' providing recommendations to industry on how AI-generated information or data should be used to support regulatory decisions regarding safety, effectiveness, or quality. This signaled that the FDA was not only adopting AI for its own processes but also establishing a framework for how sponsors should apply it in submissions.
Soon after, the FDA announced the launch of their generative AI tool, Elsa, and provided additional details on its applications. In addition to scientific review, Elsa is assisting FDA employees with reviewing clinical protocols and identifying inspection targets. Importantly, the large language model was not trained on sensitive research data.
Talks between the FDA and OpenAI to streamline drug approval
The FDA is also interested in using AI tools to help accelerate drug approval processes. In line with this, they have entered early-stage discussions with OpenAI, the organization behind ChatGPT, to explore how generative AI tools might streamline the drug approval process. These conversations have focused on identifying areas where large language models can assist with high-volume, time-intensive regulatory tasks.
The potential impact is significant. The FDA’s Center for Drug Evaluation and Research (CDER) handles tens of thousands of submissions each year, ranging from Investigational New Drug applications to post-market safety updates. The resulting project, cderGPT, is named after the center it aims to support.
Even modest efficiency gains at this scale could translate into hundreds of thousands of review hours recovered annually, compressing the overall evaluation timeline in ways that matter for sponsors managing precision medicine programs.
While still in an exploratory phase, cderGPT could support reviewers with:
- Document classification
- Summarizing trial protocols
- Flagging missing data
- Mapping applications against regulatory standards
Across all identified use cases, AI outputs would remain subject to review and oversight by FDA staff.
Implications for clinical research and precision medicine
For precision medicine programs, where submissions frequently involve biomarker-driven eligibility criteria, adaptive designs, and real-world evidence components, faster and more consistent scientific review could meaningfully reduce the regulatory lag that currently accumulates during complex evaluation cycles.
Beyond review timelines, the FDA is also integrating AI into areas directly relevant to precision medicine, including the evaluation of Digital Health Technologies and Real-World Data analytics. For sponsors building submissions around these data types, this creates an opportunity for more consistent and efficient regulatory engagement.
It is also worth noting that regulatory review is only one part of the broader timeline. Much of the delay in bringing precision therapies to patients occurs earlier, in clinical development itself: protocol design, patient identification, enrollment, and data collection. AI applied at the regulatory layer addresses one bottleneck, but the largest gains will come from applying it across the full development lifecycle.
If tools like Elsa prove reliable, developers could see fewer bottlenecks during regulatory interactions. This raises important questions about how AI-generated outputs will be documented, audited, and trusted in regulatory settings, particularly since reproducibility and explainability are essential.
Piloting internal AI assistants for study design at Sano Genetics
The FDA's adoption of AI reflects a broader shift. Across the drug development pipeline, AI-driven methodologies are already producing meaningful enhancements in both the efficiency and effectiveness of clinical processes, from target identification through post-market surveillance. At Sano Genetics, engineering teams have begun piloting internal AI assistant prototypes designed to accelerate clinical study design.
These tools aim to reduce time spent on:
- Drafting protocols from scratch
- Compiling eligibility questionnaires
- Setting up studies
In addition to speeding up initial design, the tools are also being developed to support the rapid scaling of existing studies. This can facilitate adapting studies to new regions, populations, or disease areas with minimal manual effort. While still in early development, these tools are being evaluated for their ability to reduce design cycle time, support protocol adaptation across populations and regions, and remove manual bottlenecks that currently slow study initiation in precision medicine contexts.
William Jones, Chief Technology Officer at Sano Genetics, emphasized the importance of thoughtful integration: “Sano’s philosophy has been to internally define what works best for our industry and goals. Now, we’re focused on weaving that work into our core product, first for administrators and eventually for participants, under clearly defined use cases.”
What’s next?
The FDA’s integration of AI tools like Elsa and its exploration of broader collaborations signals a shift from experimentation to application. With the 2025 draft guidance on AI in regulatory decision-making now published, the FDA has also begun formalizing expectations for how industry should use AI in submissions. As these tools and frameworks scale across regulatory workflows, they are expected to not only increase efficiency but also reshape how sponsors and regulators interact.
As these tools scale, they may enable regulatory frameworks to accommodate adaptive trial designs, real-world evidence, and larger submission volumes without extending review timelines. Human oversight remains central to the review process. AI tools are designed to handle document classification, data retrieval, and protocol summarization so that scientists can focus on safety assessment, benefit-risk analysis, and regulatory judgment.