FDA’s New Principles for AI Drug Development: Why They Matter for Bioanalysis

AI in bioanalysis is raising expectations for how bioanalytical data are generated and used in modern drug development

Artificial intelligence (AI) is becoming part of the process for discovering, developing, and testing new medicines. It is used to identify drug targets, select patients for trials, interpret complex biology, and even predict safety risks. Recognizing this shift, the FDA released in 2026 a set of ten guiding principles for the responsible use of AI in drug development. This requires a revisit of bioanalysis in this process.

U.S. and European regulators jointly agreed on guidelines for the use of AI in drug development. These 2026 principles set expectations for how AI should be designed, validated, used, and explained so that it ultimately benefits patients and earns regulatory trust. The core message rings simple: AI in drug development is welcome, but only if it is trustworthy, transparent, and used responsibly. What is less obvious, but critically important, is how these principles affect bioanalytical methods, especially those that generate PK/PD endpoint readouts in preclinical research and clinical development. In many ways, these methods sit at the foundation of AI-driven decision-making.

Although there are ten individual principles, they revolve around a few common themes:

1. Keep human oversight: AI should support human decision-making, not replace it. Scientists and clinicians must understand what AI is doing and be able to question or override it.
2. Match rigor to risk: The higher the impact of an AI-driven decision on patient safety or trial outcomes, the stronger the validation and oversight must be.
3. Be clear about purpose: AI tools must have a clearly defined context of use. An algorithm built for exploratory research should not quietly turn into a decision-making tool without additional validation.
4. Ensure data quality and traceability: AI systems are only as good as the data they learn from. Data must be well controlled, well documented, and fit for purpose.
5. Plan for the full lifecycle: AI models are not set and forget. They must be monitored, updated, and reassessed as biology, data, and clinical practice evolve.
6. Communicate clearly: Developers must clearly explain what an AI system can do, what it cannot do, and how confident users should be in its outputs.

These principles overall emphasize trust, transparency, and accountability, values that regulators already expect from traditional drug development, now applied to AI.

AI Is Becoming Infrastructure, Not a Tool

At the same time, regulators are formalizing expectations, and leading pharmaceutical companies are restructuring around AI at a systems level. In late 2025, Eli Lilly became the first pharmaceutical company to reach a US$1 trillion market capitalization and shortly thereafter announced more than $2.3 billion in AI-focused partnerships, including a co-innovation lab with NVIDIA and multiple AI-driven discovery collaborations.¹

This signals something larger than enthusiasm for new technology. It reflects a transition from using AI as a point solution to building AI-enabled development ecosystems. In this model, AI is not confined to early discovery. It spans target identification, translational modeling, dose optimization, and portfolio decision-making.

When AI becomes embedded across the development lifecycle, data are no longer isolated readouts. They fuel integrated decision systems, and under this paradigm, PK/PD data are not simply supporting documents. They are inputs into models that influence dose selection, progression decisions, and risk assessments. Variability, inconsistency, or lack of traceability at the bioanalytical level no longer stays local. It propagates through algorithms and affects strategic outcomes. This is where the FDA’s AI principles intersect directly with AI in bioanalysis.

How Modern CROs Accelerate Your Drug Development
Are you facing delays, bottlenecks, or rising costs in your drug development process? In this webinar, you will learn how leading CROs overcome these challenges with innovative strategies. Gain insights that can directly impact your timelines, reduce risks, and streamline your path to approval.
Watch the Replay

Why PK/PD Bioanalysis Suddenly Matters Even More

Pharmacokinetic (PK) and pharmacodynamic (PD) bioanalytical methods measure the amount of drug in the body and its biological effects. These data are used to answer fundamental questions, such as:

• Is the drug reaching the right tissues?
• Is it engaging its target?
• Is the biological effect dose-dependent?
• Are there early signs of safety risk or lack of efficacy?

As AI in bioanalysis becomes more common, PK/PD data are increasingly used as inputs to models, not just standalone readouts. AI systems may integrate PK/PD data with genomics, imaging, flow cytometry, or clinical observations to guide decisions about dose selection, study design, or progression to the next development stage. That means bioanalytical methods are no longer just producing numbers but feeding decision-making systems. Under the FDA’s AI principles this raises the bar for how those methods are designed, validated, and documented.

The FDA has long emphasized fit-for-purpose validation, but AI in bioanalysis makes this more visible. Bioanalytical methods must clearly state whether they are intended for:

•    exploratory signal detection,
•    mechanistic understanding,
•    dose selection, or regulatory decision support.

A PK assay suitable for early discovery may not be appropriate for driving AI-assisted dose optimization without additional rigor. AI excels at detecting patterns, but it will also amplify inconsistency. Variability caused by assay drift, reagent changes, platform upgrades, or inconsistent sample handling can be misinterpreted by AI as meaningful biology. This means stronger expectations around assay standardization, lifecycle management, and cross-study comparability of PK/PD data.

Bioanalytical Strategies Unlocked: Innovative Approaches to PK, Immunogenicity, and Biomarkers
In this webinar, our scientific experts will walk you through real-world case studies that illustrate how innovative bioanalytical approaches can help you overcome common hurdles in pharmacokinetics (PK), immunogenicity, and biomarker applications.
Watch the Replay

Under the AI principles, every data point feeding a model should be traceable. For PK/PD bioanalysis, this includes:

•    assay versions,
•    calibration strategies,
•    acceptance criteria,
•    sample preparation steps,
•    and data processing rules.

If an AI model’s output is questioned, sponsors must be able to explain the source of the PK/PD data and how it was generated.

Highly sensitive assays are not automatically “better” if the biological meaning of small changes is unclear. The FDA’s emphasis on human oversight means PK/PD data should be biologically interpretable, aligned with the mechanism of action, and presented in a way that supports scientific judgment. AI outputs that rely on PK/PD readouts must still make sense to experienced scientists.

Lifecycle thinking applies to assays, too. Just as AI models evolve, so do bioanalytical methods. Under the new principles, changes to assays that feed AI systems such as platform transitions or method refinements may require:

•    impact assessments
•    bridging studies, or re-evaluation of downstream models.

Bioanalysis is now part of a connected system rather than an isolated activity. The FDA’s AI principles do not replace existing bioanalytical guidance, but they raise expectations for how bioanalytical data are generated and used in modern drug development. Leveraging AI in bioanalysis for PK/PD means that data must be trustworthy enough for humans, robust enough for AI, and transparent enough for regulators. Where AI helps guide critical development decisions, bioanalytical methods are no longer just supporting actors; they are foundational infrastructure. Programs that recognize this early will be better positioned to move faster, make better decisions, and earn regulatory confidence along the way.

Bench + Bytes is a column written by Charles River Scientist Christoph Eberle, PhD, and Melvin Lye, Senior Director, Scientific Affairs and Product at Curiox Biosystems. It is hosted by Eureka, Charles River's scientific blog.

References:
1.    NVIDIA-Lilly Co-Innovation Lab: https://investor.lilly.com/news-releases/news-release-details/nvidia-and-lilly-announce-co-innovation-ai-lab-reinvent-drug