Using AI for drug evaluation? OpenAI is in talks with the FDA

FDA Embraces AI: Drug Approval Accelerator or Risk Exposure? According to media reports on Wednesday, the U.S. Food and Drug Administration (FDA) is in talks with OpenAI to discuss the use of artificial intelligence in drug evaluation, aiming to shorten the time for drug approval and market entry. The media cited sources familiar with the matter, stating that a small team from OpenAI has met with the FDA several times in recent weeks, along with two individuals related to the DOGE department. They are discussing a project called cderGPT, which may stand for "Center for Drug Evaluation and Research GPT." Reports indicate that FDA's first AI officer, Jeremy Walsh, is leading these discussions, but no formal contract has been signed yet. Meanwhile, on Wednesday, FDA Commissioner Marty Makary questioned on the X platform: “Why does it take more than 10 years for a new drug to hit the market? Why aren’t we modernizing with AI and other technologies? We just completed our first AI-assisted product science review, and this is just the beginning.” FDA Actively Exploring AI Applications The existing review process at the FDA typically takes about a year, but the agency has various mechanisms to accelerate the approval timeline for promising drugs, including fast track designation and breakthrough therapy designation. In recent years, the FDA has also been advancing its own research into potential uses of AI. In December 2023, the agency posted a job advertisement for researchers aimed at developing large language models for internal use. In January of this year, OpenAI announced ChatGPT Gov, a self-hosted version of the chatbot compliant with government regulations. The startup also stated that it is working to obtain FedRAMP moderate and high-level certifications for ChatGPT Enterprise, which would allow it to handle sensitive government data. Policy Guidance, Review Details, Information Authenticity... Support and Concerns Coexist Former FDA Commissioner Robert Califf told the media that the agency's review teams have been using AI for years, and it is important to focus on the specific details of AI-assisted reviews: “Understanding which parts of the review are 'AI-assisted' and what that means will be interesting.” “There has always been a demand to shorten review times, and there is a broad consensus that AI can help.” Rafael Rosengarten, CEO of precision oncology company Genialis and co-founder of the Healthcare AI Alliance, expressed support for automating certain tasks related to drug reviews, but emphasized the need for policy guidance around the types of data used to train AI models and model performance. A former FDA employee pointed out that the tendency of AI models to fabricate convincing information raises questions about the reliability of such chatbots, “Who knows how reliable this platform will be for these review tasks?” PhRMA spokesperson Andrew Powaleny believes: "It is crucial to ensure timely review of the safety and efficacy of drugs. While AI is still developing, utilizing it requires a thoughtful, risk-based approach that is patient-centered."