OpenAI, US FDA discuss using AI in drug evaluation

The US Food and Drug Administration (FDA) is in discussions with OpenAI about using AI in drug evaluation.

The talks, held in recent weeks, involve a project called “cderGPT,” potentially linked to the FDA’s Center for Drug Evaluation and Research.

Jeremy Walsh, the FDA’s first AI officer, is reportedly leading the discussions, along with representatives from OpenAI and the Department of Health and Human Services.

No formal agreements have been announced.

The FDA has explored AI use before, and experts have called for clear policies on training data and model performance. The agency aims to reduce review times using AI while maintaining safety and effectiveness.

🔗 Source: Wired

🧠 Food for thought

1️⃣ Regulatory approaches to AI in drug development show a cautious path forward

The FDA-OpenAI discussions reflect a broader regulatory evolution happening globally, with different approaches emerging between major agencies.

The FDA published discussion papers in May 2023 addressing AI/ML implications in drug development, highlighting concerns about interpreting model outputs and ensuring data quality amidst increased data exchanges 1.

While the EMA (European Medicines Agency) takes a risk-based approach categorizing AI applications based on their role in drug development, the FDA emphasizes a multidisciplinary approach with particular attention to potential “hallucinations” where AI generates inaccurate outputs 1.

These evolving frameworks explain why the FDA is proceeding carefully with initiatives like cderGPT, as they must balance innovation with reliability in a highly regulated industry.

The focus on appointing dedicated AI officers like Jeremy Walsh demonstrates how regulatory bodies are building specialized expertise to evaluate these technologies effectively rather than rushing implementation.

2️⃣ AI integration targets one bottleneck in a complex drug development pipeline

The FDA’s AI initiative addresses just one segment of a much longer process, with the typical FDA review representing only about a year in a decade-long development timeline.

Drug development costs regularly exceed $1 billion, with safety and quality assurance accounting for 50-90% of those expenses, suggesting that while AI might accelerate review processes, it addresses only a fraction of the total time and cost burden 2.