Guadalupe Hayes-Mota| Director of Bioethics, Markkula Center for Applied Ethics Santa Clara University| MIT Senior Lecturer| EU, NIH AdvisorgettySomewhere in a computational cluster, an algorithm just decided which cancer target is worth a billion dollars of research investment. Another model ranked 40,000 compounds and handed three finalists to a team of chemists. A third system is quietly influencing which patients will be enrolled in a Phase II trial scheduled to begin next month. No single human made these calls. And if the drug eventually fails or harms someone, the question of who is responsible will land on a boardroom table that was never told this was happening.This is the increasing operational reality at many of the world's largest pharmaceutical and biotech companies today. AI is now embedded in the critical path of drug discovery, and it is making consequential decisions at a speed and scale that existing governance structures were simply not designed to handle.Three High-Stakes DecisionsI see the pharmaceutical AI stack now operating across at least three distinct decision points, each carrying enormous consequences, and each usually governed by informal norms at best.The first is target selection. AI models trained on genomic, proteomic and clinical datasets are identifying which biological mechanisms to pursue. These models carry embedded assumptions about disease etiology, data quality and population representation. When a model trained predominantly on data from European ancestry cohorts ranks a target as high-confidence, that confidence score does not come with an asterisk.The second is compound prioritization. Generative chemistry models are now producing novel molecular structures and ranking them for predicted efficacy and safety. The speed is astonishing. What once took chemists years of iterative synthesis can now be proposed computationally in days. But speed is not wisdom. When a model surfaces a compound as a top candidate, the human chemists reviewing that output are operating downstream of a black box.The third is trial design and patient stratification. AI is being used to identify which patient populations will most likely respond to a therapy, and therefore, which populations get enrolled in pivotal trials. This determines not only who benefits from a drug's development but also who bears the risks of early-stage testing. When those decisions are made by algorithms trained on incomplete or historically biased datasets, the consequences are not abstract; they are borne by real patients.The Accountability VacuumAsk the chief data officer who owns an AI-selected target. They'll redirect you to the chief scientific officer. The CSO points to the model developers. The developers say they built a tool—scientists make the decisions. This circular logic is a governance structure that hasn't caught up with its own technology.No pharma company has appointed an executive solely accountable for AI-driven drug decisions. The FDA's AI framework doesn't cover discovery tools. The EMA's guidance is thoughtful but nonbinding. Billions are deployed inside a governance vacuum.Boards Are The Last Line Of DefenseBoards of pharmaceutical companies now have both the fiduciary obligation and the strategic imperative to engage with this issue directly. The fiduciary case is straightforward: Any board that has approved a digital transformation strategy without also approving an AI accountability framework has an incomplete risk picture. The liability exposure from a drug failure traceable to a flawed AI decision is not a hypothetical. It is a litigation scenario that plaintiff attorneys are already modeling.The strategic case is equally compelling. The companies that establish rigorous AI governance frameworks now can hold a durable competitive advantage in a regulatory environment. One that I believe is inevitably heading toward mandatory disclosure. When the FDA moves from guidance to requirement, the organizations with mature internal accountability structures can be well ahead of those scrambling to retrofit compliance onto opaque systems.To prepare, here is what boards should be asking right now:• Which AI systems are embedded in our critical path for target selection, compound ranking, and trial design—and who specifically owns accountability for each?• What data was used to train these models, and has that data been audited for bias, completeness and population representation?• Is there a documented process for overriding an AI recommendation, and are those overrides tracked, reviewed and escalated when patterns emerge?• How would we demonstrate, to a regulator or a jury, the decision-making chain behind a compound that caused harm?• Has our AI governance framework been reviewed by our scientific advisory board, our ethics committee and our audit committee—or only by the team that built the systems?A Framework Built For The Speed Of ScienceThe goal is not to slow AI down. The efficiency gains are real, the potential to compress drug development timelines is profound, and the promise of AI-discovered therapies for rare and neglected diseases is one of the most compelling humanitarian arguments in modern medicine. Rather, the goal is to ensure that as AI accelerates the machine of drug development, we have deliberate mechanisms for human accountability threaded through every critical junction.This means named accountability—a specific human being responsible for reviewing and signing off on each AI-assisted decision in the critical path. It means model transparency requirements with documented provenance: what data, what architecture, what validation. It means bias auditing as standard practice, with particular attention to the populations underrepresented in training data. And it also means board-level reporting and AI risk disclosure that sits alongside financial and operational risk in the governance calendar.None of this is technically difficult. But it can be organizationally difficult because it requires executives to acknowledge that their AI systems are decision-makers, and that acknowledgment carries responsibility.The algorithm that selected that cancer target is not accountable. Nor is the compound ranking model or the trial stratification system. But the board members that approved the strategy funding these systems unambiguously are. The question is whether they know it yet.​Forbes Business Council is the foremost growth and networking organization for business owners and leaders. Do I qualify?