The FDA has published draft guidance on how artificial intelligence (AI) should be evaluated in drug development, shifting focus from whether a model is called AI or machine learning to whether it's fit for purpose and supported by proper evidence. However, industry experts are raising a critical concern: while AI is accelerating drug discovery, the pharmaceutical system's ability to test and approve those discoveries remains largely unchanged, creating a dangerous mismatch in capital allocation.

What Does the FDA's New AI Guidance Actually Say?

The FDA's draft guidance establishes a practical, risk-based framework for evaluating AI credibility in drug development. Rather than treating AI and machine learning (ML) as fundamentally different technologies, the guidance focuses on context of use, model influence, and decision consequences. The key message: there is no one-size-fits-all validation strategy. The level of evidence required should be proportional to the model risk and intended use.

"The guidance encourages organizations to start with a clear definition of the question of interest and the context of use. From there, they should assess model risk based on model influence and decision consequence," explained Kishore Vatsavai, Director and Principal Scientist at Veranova.

Kishore Vatsavai, Director/Principal Scientist at Veranova

Vatsavai, who has over 20 years of experience in pharmaceutical analytical research and development, emphasized that the guidance places ultimate responsibility on people, not algorithms. "No matter how sophisticated an AI model may become, the ultimate responsibility for patient safety, product quality and regulatory compliance rests with people," he noted.

Where Is AI Creating Value in Drug Development Today?

AI is already delivering measurable benefits across multiple stages of the pharmaceutical lifecycle. In drug discovery and non-clinical development, AI is being used to predict toxicity, identify off-target effects, and prioritize promising compounds. In clinical development, it's helping improve patient stratification, optimize trial design, and support recruitment strategies. On the manufacturing and quality side, AI is applied to process monitoring, predictive maintenance, yield optimization, and deviation classification.

The applications span several key areas:

• Non-clinical Development: AI predicts toxicity and identifies off-target effects to prioritize the most promising compounds before human testing begins.
• Clinical Development: AI improves patient stratification, optimizes trial design, and supports recruitment strategies to accelerate enrollment and reduce costs.
• Manufacturing and Quality: AI monitors processes, predicts equipment failures, detects deviations sooner, and reduces batch failures to ensure reliable product quality.
• Pharmacovigilance: AI supports adverse event case triage and safety signal detection to identify potential drug safety issues faster.

What these applications have in common is that AI primarily serves as a decision-support tool. It complements scientific expertise and helps people make better, faster decisions rather than replacing human judgment.

Why Is the Industry Investing $7 Billion in AI Discovery When Zero AI-Discovered Drugs Have Been Approved?

Since January 2026, pharmaceutical companies have committed more than $7 billion to AI drug discovery partnerships, including major deals with Insilico Medicine, Servier in oncology, Eli Lilly for AI-designed oral therapeutics, SK Biopharmaceuticals in neuroimmune disease, and Takeda across multiple therapeutic areas. Yet not a single drug discovered and designed entirely by AI has received regulatory approval.

This apparent contradiction reveals where AI's proven value ends and where the next, harder problem begins. Insilico's rentosertib, the first fully AI-discovered and AI-designed molecule to publish positive Phase IIa efficacy results in a peer-reviewed study, is a genuine milestone. However, it still must navigate Phase IIb, Phase III, manufacturing validation, and full regulatory review. Every dollar committed to its discovery has bought the industry a candidate, not a drug.

The cost of converting candidates into drugs remains approximately $2.6 billion per approved product, with only about 12 percent of molecules entering clinical trials ever reaching patients. The capital allocation has not caught up to that arithmetic. The industry has spent the better part of five years building what critics call "an infinite library of beautifully bound books" while the system that determines whether anyone ever reads them remains largely unreformed.

What's the Real Bottleneck: Discovery or Development?

Industry observers argue that AI drug discovery, trained on historical molecular data, is fundamentally a more powerful filter, not a breakthrough engine. Breakthrough drugs often work because they do something unexpected in a living human system. Train a model on prior success and you get a system biased toward consensus, one that is constitutionally incapable of producing biological breakthroughs. The real uncertainty in drug development was never whether chemists could make an interesting molecule. It was always whether that molecule would survive contact with human biology, a regulatory agency, and a clinical trial infrastructure running at 80 percent enrollment delay rates and costing sponsors $55,000 per day in late-stage delays.

Clinical development has remained AI-resistant even as discovery has not, for two structural reasons. First, pharmaceutical companies hold decades of trial data that is neither standardized nor model-ready. The population-scale sources that could give AI genuine predictive power in development, such as biobanks, real-world evidence repositories, and integrated electronic health record networks, represent a second data layer that current foundation model platforms do not yet reach. Sponsors who have run 50 Phase II trials do not automatically have 50 trials' worth of usable training data. They have 50 silos.

Second, there is a trust sequencing problem. In clinical development, a failed pivotal trial can wipe hundreds of millions in pipeline value in a single readout. New tools earn trust in low-stakes pilot settings first. This is not conservatism for its own sake. It is a rational institutional response to a failure mode with asymmetric consequences. The enterprise-deployment model that governs most software adoption does not translate to an environment where the cost of a wrong decision is not a bad quarter but a dead program.

How Should Pharma Companies Prepare for AI in Drug Development?

Industry experts recommend a phased approach to AI adoption in pharmaceutical development, focusing on practical, portfolio-level applications rather than single-asset bets:

• Start with Clinical Operations: Enrollment forecasting, site optimization, and workflow automation are not moonshots. They are infrastructure plays that compound across every program in a portfolio. A sponsor running 15 concurrent Phase II trials does not need 15 separate AI discoveries. It needs one enrollment prediction model that works reliably across all 15.
• Prioritize Data Standardization: Before AI can contribute meaningfully to clinical decision-making, pharmaceutical companies must standardize decades of trial data and integrate it with external sources like biobanks and real-world evidence repositories. This is foundational work that cannot be skipped.
• Sequence Adoption Pilots: Deploy AI tools in low-stakes pilot settings first to build institutional trust. The FDA has not yet issued guidance permitting an AI-generated clinical recommendation to substitute for investigator judgment in a GCP-regulated trial, so the gap between "useful workflow tool" and "accountable clinical decision support" remains wide.
• Mitigate Bias and Model Drift: Bias mitigation and model drift are among the most important considerations when deploying AI in regulated pharmaceutical environments. Even a highly accurate model can become unreliable if it drifts from its original training conditions or if underlying data distributions change.

The FDA's draft guidance preserves a flexible, risk-based framework while providing clarity on implementation. However, industry observers are looking for greater clarity on documentation requirements for different categories of AI applications, lifecycle management for adaptive or continuously learning models, and accountability when models, data sources, or underlying technologies are provided by external providers.

What Does This Mean for the Future of Drug Approval?

The counterintuitive read on the $7 billion committed to AI drug discovery is that it validates clinical development AI more than it validates discovery AI. Pharma companies have now publicly confirmed, through their own capital allocation, that they believe AI can find better molecules faster. If that belief is correct, the pipeline pressure flowing toward clinical development will intensify, not ease. More candidates entering Phase I means more candidates competing for site capacity, for enrollment bandwidth, for regulatory review time. A faster discovery engine attached to an unreformed development engine does not accelerate drug approval. It accelerates congestion.

The question the industry has not yet answered at scale is this: if the industry is willing to invest $7 billion to discover better molecules faster, when will it invest at that scale to help those molecules reach patients faster? Rentosertib will provide a partial answer as it moves through the development machine. But the machine itself has not changed.