Artificial Intelligence, Virtual Participants, and the Evolving FDA Regulatory Landscape

Artificial intelligence is moving quickly from an experimental capability to a foundational part of how clinical research is designed, executed, analyzed, and regulated. Regulatory agencies, including the U.S. Food and Drug Administration, are responding in parallel. Over the last several years, the FDA has expanded its focus on computational modeling, simulation, machine learning, and AI-enabled evidence generation across both drug and device development.¹

Practical pressures across the clinical trial ecosystem are driving this shift. Development timelines remain long, recruitment challenges continue to slow studies, protocol complexity is increasing, and sponsors are under growing pressure to generate higher quality evidence more efficiently. AI and computational modeling offer potential ways to improve planning, accelerate decision-making, and explore trial outcomes before patients are enrolled.

The FDA has shown growing support for these approaches, particularly when used for trial design optimization, simulation, feasibility analysis, and supplementary evidence generation. At the same time, the agency has made clear that AI-generated outputs must meet rigorous standards for transparency, validation, governance, and credibility. Context of use matters. Risk level matters. Documentation matters.

For sponsors, CROs, and technology providers, the conversation is no longer centered on whether AI will play a role in regulated clinical research. The focus is increasingly on how organizations implement AI responsibly and how they prepare for evolving regulatory expectations around oversight, reproducibility, and evidence quality.

This paper examines how FDA thinking around AI and virtual participant models has evolved, what regulatory frameworks currently exist, what expectations researchers should anticipate, and what organizations should consider when building an AI-ready clinical research strategy.

Why AI Has Become a Regulatory Imperative

Clinical development continues to face the same pressures that have challenged the industry for years, but those pressures are becoming harder to absorb through traditional operational improvements alone. Studies are more complex, timelines remain difficult to compress, and sponsors are expected to generate increasingly robust evidence while controlling cost and risk.

Protocol complexity has grown steadily over the last decade.² More endpoints, more procedures, more data streams, and more decentralized elements have introduced operational strain across nearly every phase of development. Recruitment and retention remain persistent obstacles, particularly in long-duration studies and specialized therapeutic areas. At the same time, regulators and payers are demanding stronger evidence packages that reflect broader patient populations and real-world variability.

AI has entered this environment as both an opportunity and a governance challenge.

Regulators are paying attention because AI systems are now influencing decisions across healthcare and life sciences at a meaningful scale. Machine learning models are being used to identify patients, optimize protocols, forecast enrollment, monitor safety signals, analyze imaging data, and support operational planning. Computational modeling and simulation approaches are also becoming more sophisticated, including the emergence of synthetic control arms and virtual participant models capable of simulating long-term clinical outcomes.

These technologies offer potential advantages that are difficult to ignore. Researchers can evaluate multiple trial scenarios before enrollment begins, explore protocol tradeoffs earlier, and model long-term outcomes in compressed timeframes. AI-assisted approaches may also reduce unnecessary operational burden by helping teams identify risks and inefficiencies earlier in the development process.

The FDA has acknowledged these opportunities publicly through a growing body of guidance, discussion papers, and regulatory science initiatives focused on AI and computational modeling. At the same time, the agency has consistently emphasized that innovation alone is not sufficient for regulatory acceptance. AI-generated outputs must be credible, transparent, validated, and appropriate for their intended use.

This tension between innovation and oversight now defines much of the regulatory conversation surrounding AI in clinical research. The industry is moving quickly toward broader adoption, while regulators are working to establish frameworks that preserve scientific rigor, patient safety, and confidence in the evidence being generated.

How FDA Thinking Around AI and Computational Evidence Has Evolved

The FDA’s position on AI and computational evidence has evolved significantly over the last several years.³ Early discussions around computational modeling were often confined to niche applications, particularly in medical device development and engineering simulations. Today, AI and modeling technologies are being discussed much more broadly as part of the future infrastructure of drug development and regulatory science.

This shift reflects both technological progress and practical necessity. The volume and complexity of clinical data have expanded rapidly, creating pressure for more scalable approaches to analysis, forecasting, and evidence generation. Regulators recognize that traditional methods alone may not be sufficient to support the pace and complexity of modern development programs.

The FDA has increasingly positioned computational modeling an d simulation as valuable regulatory science tools.⁴ Programs involving in silico methods, model-informed drug development, and digital health technologies have all contributed to this broader acceptance. The agency has also invested in internal expertise and public-private collaboration efforts aimed at understanding how AI systems should be evaluated and governed in regulated environments.

One of the clearest changes has been the movement from theoretical discussion to operational expectations. Earlier conversations focused heavily on AI’s future potential. More recent FDA communications focus on oversight, credibility assessment, lifecycle management, transparency, and risk-based governance.

The rise of in silico clinical trial methodologies illustrates this progression clearly. These approaches use computational models to simulate patient responses, disease progression, or treatment outcomes under defined conditions. In some settings, they can reduce the need for extensive physical testing or support decision-making before large-scale enrollment begins.

The FDA has shown particular openness to simulation-based approaches in areas such as:

medical device evaluation
protocol optimization
dose modeling
statistical forecasting
supplementary evidence generation

At the same time, the agency has maintained important boundaries around where AI-generated evidence can currently substitute for biological evidence derived from human participants.

Another notable evolution is the FDA’s increasing focus on context of use. Regulators now place significant emphasis on defining exactly how an AI model will be used and how much regulatory weight will depend on its outputs. A model used for operational planning may face lower scrutiny than one supporting efficacy conclusions or safety-related decisions.

This framework allows regulators to encourage innovation while maintaining proportional oversight. It also signals that organizations deploying AI systems must think carefully about validation strategies, governance structures, and intended applications from the beginning rather than treating them as secondary considerations later in development.

The FDA’s Current Position on AI-Generated and Synthetic Clinical Evidence

The FDA’s current position on AI-generated evidence can best be described as supportive, conditional, and risk-based.¹ The agency recognizes the value of AI and computational modeling while continuing to draw clear distinctions between supplementary analytical tools and primary evidence required for regulatory approval.

In practical terms, the FDA has shown growing support for AI-assisted approaches that improve trial planning, operational efficiency, and evidence generation processes. Computational modeling and simulation are increasingly viewed as legitimate components of modern development programs when used within clearly defined boundaries.

Areas where regulators have demonstrated meaningful openness include:⁵

trial design optimization
protocol feasibility analysis
statistical power modeling
patient population simulation
dose-response modeling
long-term outcome forecasting
synthetic control arm supplementation

These applications are attractive because they may improve efficiency and help sponsors make better-informed decisions earlier in development. They also align with broader regulatory goals around modernizing clinical research and reducing unnecessary burden where scientifically appropriate.

However, important limitations remain.

For most drug and biologic approvals, the FDA still requires evidence generated from real human participants to establish safety and efficacy. Virtual participants and synthetic datasets are not currently recognized as wholesale replacements for pivotal human clinical trial evidence in standard approval pathways.

This distinction is central to understanding the current regulatory environment.

The FDA’s approach is heavily influenced by what is commonly referred to as a fit-for-purpose credibility framework. Under this philosophy, acceptance depends on several interrelated factors:

the model’s intended use
the potential regulatory impact
the level of risk associated with incorrect outputs
the strength of validation evidence
the transparency of the methodology

A simulation model used to explore operational scenarios may face relatively limited scrutiny compared to a model supporting efficacy conclusions tied directly to approval decisions.

Context of use has therefore become one of the most important concepts in regulatory AI discussions. Sponsors and technology developers are increasingly expected to define:

what the system does
how it will be used
what decisions depend on it
what limitations exist
how outputs are validated

The FDA has also emphasized transparency repeatedly. Black-box systems with limited explainability or weak documentation are likely to face greater skepticism, particularly in higher-risk applications. Regulators want to understand how models were trained, what data sources were used, how performance was evaluated, and where limitations may exist.

This does not mean regulators expect perfect predictability from AI systems. Clinical research inherently involves uncertainty and variability. The expectation is that organizations demonstrate credible governance, appropriate validation, and scientific rigor proportional to the role AI plays within the broader evidence package.

Existing Regulatory Frameworks and Guidance Researchers Should Understand

The regulatory landscape surrounding AI in clinical research remains evolving, but several themes and frameworks are already emerging clearly. Organizations building or deploying AI-enabled systems should understand these expectations now rather than waiting for fully mature regulations to arrive.

One major theme is lifecycle oversight. FDA discussions around AI and machine learning consistently emphasize that oversight cannot stop at deployment.⁶ Models may evolve over time as data changes, performance drifts, or operating conditions shift. Regulators increasingly expect organizations to monitor, document, and govern AI systems continuously rather than treating validation as a one-time event.

Another important concept is risk-based regulation. The FDA has repeatedly signaled that regulatory expectations should align with the level of risk associated with a system’s intended use. Lower-risk operational tools may face lighter scrutiny than systems directly influencing patient safety, efficacy assessments, or regulatory decision-making.

Computational modeling and simulation frameworks are also becoming more influential.⁷ The FDA and broader scientific community have increasingly referenced credibility assessment approaches such as the ASME V&V 40 framework, which focuses on verification, validation, and fitness for purpose in computational modeling. These frameworks help establish structured approaches for evaluating whether a model is sufficiently reliable for its intended context.

Transparency remains another consistent regulatory priority.

Organizations using AI-enabled systems should expect growing expectations around:

training data documentation
data provenance
reproducibility
auditability
version control
performance monitoring
change management

These expectations reflect broader concerns about reliability, bias, and explainability. Regulators want organizations to demonstrate that AI outputs can be understood, traced, and evaluated within a governed quality framework.

Governance structures themselves are becoming increasingly important. AI oversight is no longer viewed solely as a technical function. Effective governance often requires collaboration across clinical operations, biostatistics, regulatory affairs, quality, data science, and compliance teams.

Researchers should also anticipate increasing scrutiny around data quality. AI systems are highly dependent on the quality, representativeness, and consistency of the datasets used to train and validate them. Incomplete, biased, or poorly curated data can introduce significant reliability concerns that directly affect regulatory confidence.

Another important trend is the growing distinction between assistive AI and autonomous AI. Systems that support human decision-making while preserving meaningful human oversight may face fewer barriers than systems operating with limited transparency or minimal human review.

Taken together, these frameworks signal a broader regulatory direction. The FDA is not discouraging AI adoption. The agency is moving toward structured expectations that allow innovation to scale within a framework of scientific accountability, documentation, and risk management.

What Sponsors and Researchers Should Expect From AI-Enabled Systems

Organizations implementing AI in clinical research should expect a regulatory environment that increasingly focuses on evidence quality, governance maturity, and operational accountability. The discussion is moving beyond whether AI tools can improve efficiency. Regulators now want to understand whether these systems are reliable, transparent, and appropriate for regulated decision-making environments.

Validation sits at the center of these expectations.

AI-generated outputs are unlikely to gain meaningful regulatory acceptance without evidence demonstrating that the models perform consistently and credibly under real-world conditions. Sponsors should expect increasing pressure to benchmark models against biological, clinical, or operational datasets relevant to the intended application.

Validation may involve:

retrospective comparisons
prospective testing
simulation-to-observed outcome analysis
sensitivity analysis
reproducibility assessments

The level of rigor required will depend heavily on the role the system plays within the broader development process.

Researchers should also expect growing scrutiny around explainability and transparency.⁸ Highly opaque models may create challenges in regulated environments where stakeholders need to understand how conclusions were generated and where limitations exist. Explainability requirements will vary by context, but organizations should assume that documentation and interpretability will remain important themes in regulatory review.

Data quality is likely to become one of the most significant operational differentiators in AI-enabled clinical research. AI systems inherit many of the weaknesses present in their underlying data sources. Inconsistent source data, limited population diversity, missing variables, or poorly structured datasets can reduce reliability and increase bias risk.

As a result, sponsors should expect increasing emphasis on:

data lineage
dataset representativeness
metadata management
standardization
traceability

Governance expectations are also expanding quickly.⁹ Regulators increasingly view AI oversight as an organizational capability rather than a technical feature. Effective governance requires defined accountability structures, clear operating procedures, change management controls, and ongoing monitoring practices.¹⁰

Human oversight remains especially important. Current regulatory thinking generally favors systems that augment expert judgment rather than replace it entirely. Researchers should expect regulators to examine how humans review, interpret, and challenge AI-generated outputs within operational workflows.

Organizations that approach AI implementation as a governed scientific process rather than a standalone technology deployment will likely be better positioned as regulatory expectations continue to mature.

Building an AI-Ready Regulatory and Clinical Strategy

Many organizations are still approaching AI adoption tactically, focusing on isolated tools or narrow operational efficiencies. That approach may deliver short-term gains, but it often creates fragmented oversight, inconsistent validation practices, and governance gaps that become more problematic as AI usage expands.

An AI-ready strategy requires a more structured foundation.

The starting point should be clearly defined use cases.¹¹ Broad or loosely defined AI initiatives can create unnecessary regulatory and operational risk. Organizations should identify specific problems where AI can provide measurable value and where outputs can be validated against meaningful benchmarks.

Examples may include:

protocol optimization
enrollment forecasting
synthetic cohort modeling
operational risk prediction
endpoint trend analysis
long-term simulation studies

Defining intended use early is particularly important because regulatory expectations will depend heavily on how the system influences decision-making.

Organizations should also align validation rigor with regulatory impact. Systems supporting high-risk decisions require stronger evidence, deeper documentation, and more robust oversight structures. Applying proportional governance allows organizations to scale AI responsibly without overengineering lower-risk applications.

Early regulatory engagement can also play an important role. Sponsors developing novel AI-enabled methodologies may benefit from pre-IND discussions, Q-submission pathways, or other early interactions with regulators. These conversations can help clarify expectations, reduce uncertainty, and identify potential concerns before major investments are made.

Infrastructure decisions matter as well.

AI-ready environments increasingly require systems capable of supporting:¹²

auditability
reproducibility
model traceability
controlled workflows
version management
governed data access
longitudinal monitoring

Organizations should also prepare for ongoing regulatory evolution. Current guidance represents an intermediate stage rather than a finalized framework. Expectations surrounding AI governance, synthetic evidence, and computational modeling will likely continue to expand over the coming years.

Cross-functional alignment is becoming increasingly important in this environment. AI strategy can no longer sit entirely within isolated innovation or data science teams. Regulatory affairs, quality, clinical operations, biostatistics, compliance, and technology leadership all play interconnected roles in establishing credible governance structures.

The organizations most likely to succeed will not necessarily be the ones adopting AI fastest. They will be the ones building scalable oversight frameworks capable of supporting innovation while maintaining scientific rigor, operational transparency, and regulatory confidence.

Looking Ahead: Where the Regulatory Environment Is Heading

The regulatory environment surrounding AI in clinical research is likely to become both more structured and more accepting over time. Current FDA activity¹ suggests that regulators see AI and computational modeling as increasingly important components of modern development infrastructure rather than temporary or experimental technologies.

Acceptance will probably expand gradually and unevenly across use cases.

Lower-risk applications tied to operational planning, simulation, and supplementary evidence generation are likely to continue gaining traction first. These applications allow regulators and industry stakeholders to build confidence in AI-enabled methodologies while maintaining traditional evidence standards for higher-risk approval decisions.

More advanced uses involving synthetic participants, virtual cohorts, and predictive simulation will likely continue evolving as validation methods improve and larger bodies of supporting evidence emerge. Some therapeutic areas and development contexts may adopt these approaches more quickly than others, particularly where recruitment challenges, rare diseases, or ethical constraints create limitations for traditional study designs.

At the same time, full replacement of human participants in pivotal trials remains unlikely in the near term for most drug and biologic approvals. Scientific variability, biological complexity, and patient safety considerations continue to make real-world human evidence central to regulatory decision-making.

The future will likely be hybrid rather than fully virtual.

Clinical development programs may increasingly combine:

biological participants
synthetic control arms
AI-assisted operational planning
simulation-based forecasting
computational evidence generation

This blended model could help improve efficiency while preserving confidence in safety and efficacy conclusions grounded in human data.

Organizations should also expect AI governance requirements to become more formalized. Expectations around documentation, lifecycle oversight, reproducibility, transparency, and performance monitoring will likely continue expanding as AI systems become more deeply embedded in regulated workflows.

The broader competitive landscape may shift as well. Sponsors and technology providers capable of demonstrating governed, validated, and traceable AI workflows could gain advantages in operational efficiency, regulatory confidence, and development scalability.

AI is increasingly becoming part of the operational foundation of clinical research. The organizations best positioned for the next phase of this transition will likely be those that treat governance, validation, and infrastructure readiness as strategic priorities rather than compliance afterthoughts.

Conclusion

AI is steadily becoming part of the operational and regulatory foundation of clinical research. Regulatory agencies are no longer treating computational modeling, simulation, and AI-assisted evidence generation as fringe concepts. The focus has shifted toward defining how these technologies can be applied responsibly within frameworks that preserve scientific rigor, transparency, and patient safety.

For sponsors, CROs, and research organizations, this creates both opportunity and pressure. AI may help accelerate study startup, improve operational planning, support simulation-based forecasting, and reduce inefficiencies across the trial lifecycle. At the same time, organizations are being asked to think more carefully about validation, governance, traceability, and oversight as these technologies become more deeply integrated into regulated workflows.

The organizations best positioned for the next phase of clinical development will likely be those that approach AI strategically rather than experimentally. That means building infrastructure capable of supporting governed AI workflows, maintaining clear documentation and auditability, and aligning innovation efforts with evolving regulatory expectations from the beginning.

Crucial Data Solutions and the TrialKit platform help research teams integrate AI into clinical studies in ways that are practical, scalable, and aligned with modern regulatory and operational expectations. From study design and workflow support to simulation and structured data oversight, TrialKit is designed to help organizations apply AI capabilities effectively within real-world research environments.

Learn more about how TrialKit AI can support your clinical development strategy or contact the Crucial Data Solutions team to schedule a discussion.

¹ “Artificial Intelligence for Drug Development.” FDA, https://www.fda.gov/about-fda/center-drug-evaluation-and-research-cder/artificial-intelligence-drug-development.

² McGowan, Brian S. “Do Phase II Protocol Learnings Help Simplify Phase III Protocol Designs?” Applied Clinical Trials, 13 October 2025, https://www.appliedclinicaltrialsonline.com/view/do-phase-ii-protocol-learnings-help-simplify-phase-iii-protocol-designs-?

³ “Using Artificial Intelligence & Machine Learning in the Development of Drug and Biological Products.” FDA, https://www.fda.gov/media/167973/download?

⁴ “Model-Informed Drug Development Paired Meeting Program.” FDA, 15 January 2026, https://www.fda.gov/drugs/development-resources/model-informed-drug-development-paired-meeting-program?

⁵ “Assessing the Credibility of Computational Modeling and Simulation.” U.S. Food and Drug Administration, 2023, www.fda.gov/regulatory-information/search-fda-guidance-documents/assessing-credibility-computational-modeling-and-simulation-medical-device-submissions?

⁶ “Good Machine Learning Practice for Medical Device Development.” FDA, 19 December 2025, https://www.fda.gov/medical-devices/software-medical-device-samd/good-machine-learning-practice-medical-device-development-guiding-principles?

⁷ “AI Risk Management Framework | NIST.” National Institute of Standards and Technology, https://www.nist.gov/itl/ai-risk-management-framework.

⁸ “Transparency for Machine Learning-Enabled Medical Devices.” U.S. Food and Drug Administration, 2024, www.fda.gov/medical-devices/software-medical-device-samd/transparency-machine-learning-enabled-medical-devices-guiding-principles?

⁹ “Reflection paper on the use of Artificial Intelligence (AI) in the medicinal product lifecycle_240903.” EMA, 9 September 2024, https://www.ema.europa.eu/en/documents/scientific-guideline/reflection-paper-use-artificial-intelligence-ai-medicinal-product-lifecycle_en.pdf.

¹⁰ “GUIDELINE FOR GOOD CLINICAL PRACTICE E6(R3).” ICH, 6 January 2025, https://database.ich.org/sites/default/files/ICH_E6%28R3%29_Step4_FinalGuideline_2025_0106.pdf.

¹¹ “GENERAL CONSIDERATIONS FOR CLINICAL STUDIES E8(R1).” ICH, 6 October 2021, https://database.ich.org/sites/default/files/ICH_E8-R1_Guideline_Step4_2021_1006.pdf.

¹² “Digital Health Technologies (DHTs) for Drug Development.” FDA, 29 Mar. 2023, www.fda.gov/science-research/science-and-research-special-topics/digital-health-technologies-dhts-drug-development.