FDA proposes risk-based framework for AI in drug development

The guidance covers AI intended to support regulatory decisions about a drug or biological product’s safety, effectiveness or quality.[1] The FDA says the framework is designed to help sponsors assess and establish model credibility for a particular context of use, rather than treating a model as credible in the abstract.[1] The agency defines context of use as how AI is used to address a particular question of interest.[1] The FDA said it developed the draft with input from sponsors, manufacturers, technology developers and suppliers, and academics.[1]

The draft proposes a risk-based framework for evaluating whether an AI model is credible in a specific context.[2] Under the FDA’s definition, context of use connects the model to the particular question it is being used to address.[1] The FDA describes credibility as trust in a model’s performance for that specific context.[1] This makes the intended question—not simply the model itself—the focus of the proposed assessment.[2][1]

Clinical Trial Vanguard editor Moe Alsumidaie argued that sponsors bear the documentation burden under the draft framework.[3] He said sponsors can show the FDA how they assessed a model’s fitness for a specific use before using it.[3] Alsumidaie identified risk-based monitoring, anomaly detection and data-quality flagging as oversight functions where sponsors have moved quickly to deploy AI.[3] He also reported that Phase III trials collected an average of 5.9 million data points per protocol, up 11% annually since 2020.[3] These comments frame the framework as relevant to teams assessing AI for trial oversight.[1][3]

The FDA says the recommendations are nonbinding and the draft is not for implementation.[2] It sought public comment within 90 days and encouraged sponsors to engage early about credibility assessments or AI use in human and animal drug development.[1] Alsumidaie also reported that an FDA warning letter issued in April 2026 was widely reported as the agency’s first to explicitly cite AI misuse.[3] He noted that the warning letter concerned manufacturing, not clinical trials.[3] That distinction matters when considering the letter alongside guidance focused on drug-development uses.[1][3]