By Editorial Team 
The cautionary tale of Theranos, culminating in a belated FDA warning letter, offers a stark blueprint for the critical need to proactively address AI safety in healthcare. The question for regulators, innovators, and investors alike is not if such failures will recur with AI, but how we can prevent them from reaching a similar scale of patient and economic impact.
The Theranos Precedent: A Failure of Proactive Oversight
The Theranos saga, characterized by ambitious claims and a profound lack of scientific rigor, serves as a potent reminder of the dangers inherent in medical technology operating outside robust regulatory scrutiny. While Theranos primarily dealt with blood diagnostics rather than AI, the core issues of unvalidated technology, opaque operations, and the potential for widespread patient harm resonate deeply with emerging concerns in AI health. The FDA’s delayed intervention, ultimately manifesting in a warning letter, came after years of the company operating with significant public and investor confidence. This delay allowed unproven technologies to potentially influence patient care, creating a substantial covered-lives impact and ultimately leading to significant economic repercussions for investors and the healthcare system. The trajectory of Theranos underscores a critical relationship: FDA’s delayed intervention on Theranos demonstrates the need for proactive AI safety oversight. Had the FDA CDRH, perhaps under leadership like that of Bakul Patel during his tenure, been empowered or structured to intervene earlier and more decisively, the fallout could have been mitigated. Patel is now Senior Director, Global Digital Health Strategy & Regulatory at Google. This isn’t merely a historical critique; it’s a forward-looking imperative for AI. As Eric Topol has frequently articulated, the promise of AI in healthcare is immense, but so too is the peril if validation and oversight lag behind development. The economic impact of delayed oversight is clear: billions in lost investment and eroded public trust, directly attributable to a failure to rigorously vet claims and ensure foundational safety.
Lessons for AI: Preventing the Next “Theranos” Through Proactive Regulation
For the AI health sector, the lessons from Theranos are not merely analogous; they are prescriptive. The rapid development cycle of AI, particularly in Software as a Medical Device (SaMD), necessitates a regulatory approach that is both agile and anticipatory. Unlike traditional medical devices, AI models are dynamic; they can exhibit algorithmic drift, and their performance can degrade over time as real-world data distributions shift away from their training data. Without a Predetermined Change Control Plan (PCCP) and continuous monitoring, every update could theoretically introduce new risks, demanding a new 510(k) clearance, an unscalable proposition. Investors and venture capitalists, keenly aware of exit multiples and reimbursement pathway clarity, must integrate rigorous AI safety diligence into their investment theses. The absence of robust quality management systems (QMS), particularly adherence to standards like ISO 13485, or a clear strategy for real-world evidence (RWE) generation post-deployment, should be considered significant red flags. Just as Theranos lacked the fundamental scientific validation, an AI company without a clear roadmap for monitoring and managing model performance in real-world settings presents an equivalent, if not greater, risk. The economic cost of regulatory missteps or eventual product recalls in the AI space could be substantial, impacting not just individual companies but potentially chilling investment across the entire sector.
The Role of Regulatory Frameworks and Enforcement
The existing regulatory landscape provides a foundation, but it needs to evolve. The FDA SaMD Framework offers guidance for AI/ML-based medical devices, emphasizing a total product lifecycle approach. However, the Theranos case highlights that guidance alone is insufficient without robust enforcement mechanisms. The FDA 510(k) Pathway, while a common route for many AI products demonstrating substantial equivalence, may not fully address the unique challenges of continuously learning algorithms. Novel AI applications, requiring De Novo classification, face more stringent review, but even these initial clearances do not guarantee sustained performance or safety. The Department of Justice (DOJ) involvement in the Theranos case underscores the ultimate consequences of systemic failures in regulatory compliance and ethical conduct. For AI, this translates to a need for clear accountability frameworks. As Scott Gottlieb, a former FDA Commissioner, has emphasized, maintaining public trust in medical innovation requires unwavering commitment to safety and efficacy. This means not only scrutinizing pre-market claims but also establishing robust post-market surveillance for AI. The economic implications of an AI system failing in the field, from misdiagnoses leading to adverse patient outcomes to the financial burden on healthcare systems and potential litigation, are immense. FDA guidance on post-market surveillance for AI/ML medical devices
Shaping the Future of AI Health Oversight
The critical takeaway from the Theranos experience for AI in healthcare is the urgent need for a paradigm shift from reactive enforcement to proactive safety oversight. This requires a collaborative effort between regulators, innovators, and investors. For regulatory bodies like FDA CDRH, this means developing more dynamic and adaptive regulatory tools, potentially leveraging principles of Good Machine Learning Practice (GMLP) as a baseline for development and deployment. It also means fostering an environment where concerns about AI health misinformation and potential failures are addressed swiftly and transparently, preventing them from escalating into crises of trust and patient safety. Investors, in turn, must demand not just compelling technological breakthroughs but also demonstrably safe and clinically validated AI solutions. Due diligence should extend beyond the immediate market opportunity to encompass a company’s long-term strategy for regulatory compliance, data governance (including adherence to HIPAA, HITRUST, and SOC 2), and continuous model validation. The economic impact of an AI health failure, particularly one that undermines public trust, can be far-reaching, affecting reimbursement pathways and market adoption. By prioritizing robust safety frameworks from inception, the industry can ensure that AI’s transformative potential is realized responsibly, safeguarding both patient well-being and the integrity of the healthcare investment landscape. Commonwealth Fund report on health equity and AI The specter of Theranos should not stifle innovation, but rather sharpen our focus on building a future where AI in healthcare is synonymous with both breakthrough and unwavering safety. Eric Topol’s publications on AI in medicine
Frequently Asked Questions
For FDA/Regulatory Officers: How does the Theranos case inform the FDA’s approach to AI safety, particularly regarding proactive oversight?
The Theranos case highlights the need for proactive AI safety oversight, as the FDA’s delayed intervention allowed unproven technologies to influence patient care and caused significant economic repercussions. For AI, this means developing more dynamic and adaptive regulatory tools and addressing concerns swiftly to prevent crises of trust and safety.
For FDA/Regulatory Officers: What specific regulatory challenges does AI in healthcare present that were not as prominent with traditional medical devices, as illustrated by Theranos?
AI models are dynamic and can exhibit algorithmic drift, meaning their performance can degrade over time. This necessitates a regulatory approach that is both agile and anticipatory, unlike traditional medical devices, and requires continuous monitoring and potentially Predetermined Change Control Plans (PCCP).
For Investors/VCs: What red flags, similar to those in the Theranos case, should investors look for when evaluating AI health companies?
Investors should consider the absence of robust quality management systems (QMS), particularly adherence to standards like ISO 13485, or a clear strategy for real-world evidence (RWE) generation post-deployment as significant red flags. An AI company without a clear roadmap for monitoring and managing model performance in real-world settings presents a substantial risk.
For Investors/VCs: What are the potential economic impacts of regulatory missteps or product recalls in the AI health sector, drawing parallels to Theranos?
The economic cost of regulatory missteps or eventual product recalls in the AI space could be substantial, impacting not just individual companies but potentially chilling investment across the entire sector. The Theranos case resulted in billions in lost investment and eroded public trust due to a failure to rigorously vet claims and ensure foundational safety.
