The Billion Dollar Graveyard: AI’s Clinical Validation Crisis

The promise of artificial intelligence in healthcare is transformative, yet a sobering reality exists beneath the hype: a graveyard of once-lauded companies, representing over $62 billion in combined value destruction. What unites these fallen giants, from Theranos’s spectacular implosion to the quiet struggles of others? The answer, time and again, points to fundamental safety failures and a critical disconnect from the rigorous clinical validation demanded by healthcare.

The Mirage of Unvalidated Innovation: Lessons from Theranos to Olive AI

The narrative of unchecked ambition overriding clinical necessity is perhaps best exemplified by Theranos. While not an AI company in the modern sense, its downfall serves as a stark precursor, demonstrating the catastrophic consequences when claims of revolutionary technology are not underpinned by verifiable, reproducible patient safety data. The company’s promises of ubiquitous, low-cost blood testing crumbled under scrutiny, revealing a profound lack of clinical validation and ultimately, patient harm. This foundational lesson, that innovation must be proven safe and effective, echoes through the more recent struggles of AI-driven ventures. Consider Olive AI, once valued at over $4 billion, which aimed to automate administrative tasks in healthcare. Despite significant investment, its solutions often failed to deliver the promised efficiencies and, in some instances, introduced new complexities or errors, demonstrating a failure to integrate safely and effectively into existing clinical workflows. The company ultimately shut down in November 2023 after selling off its core business units. This highlights a common pitfall: AI solutions, even those not directly involved in diagnosis or treatment, can introduce safety risks if their performance is not rigorously assessed in real-world healthcare settings. The challenges faced by Babylon Health, which pivoted from a direct-to-consumer AI-powered symptom checker to a more traditional healthcare provider model, further underscore this point. Babylon Health ceased global operations by September 2023, with its US operations closed and UK operations sold and rebranded. Its initial offerings, while innovative, faced persistent questions regarding the accuracy and safety of its AI-driven diagnostic guidance (DP12), raising concerns among patient safety advocates.

The Peril of Premature Commercialization: Pear Therapeutics and Proteus Digital Health

The journey from promising technology to sustainable, safe healthcare solution is fraught with regulatory and clinical hurdles. Pear Therapeutics, a pioneer in prescription digital therapeutics (PDTs), achieved significant regulatory milestones, including FDA clearances. However, despite these clearances, the company ultimately declared bankruptcy in April 2023 and finished winding down two years later in 2026. This outcome, while complex, suggests that regulatory approval alone, even within the FDA SaMD Framework, does not guarantee commercial viability or widespread clinical adoption if the real-world impact on patient outcomes and safety isn’t compelling enough for payers and providers. Similarly, Proteus Digital Health, with its ingestible sensors designed to monitor medication adherence, faced significant challenges despite FDA approval, filing for Chapter 11 bankruptcy in June 2020 and subsequently being acquired by Otsuka Pharmaceutical. The technology, while ingenious, struggled with integration into clinical practice and demonstrating clear, consistent patient safety benefits and cost-effectiveness at scale. These cases illustrate that even when a technology passes initial regulatory muster, its long-term success hinges on demonstrable clinical utility and a clear, positive impact on patient safety and health outcomes. Without this, even groundbreaking innovation can become part of the $62B+ in combined value destruction.

The Unseen Risks: Data Integrity and Clinical Oversight

The rise and fall of companies like Cerebral, a mental health platform that faced scrutiny over prescribing practices and cut services and laid off staff in 2024-2025, and the ongoing challenges at Teladoc-Livongo, highlight the critical role of human oversight and data integrity in AI-driven healthcare. Teladoc Health recorded a $13.7 billion write-down related to the Livongo acquisition, and its CEO stepped down in April 2024. While AI can augment clinical decision-making, it cannot replace responsible human judgment, especially when patient safety is at stake. Cerebral’s rapid expansion, coupled with allegations of inappropriate prescribing, underscores the dangers of scaling healthcare AI without robust clinical governance and safety protocols. Even companies focused on genomics, such as 23andMe and Ambry Genetics, have faced their own set of challenges. 23andMe experienced a significant data breach in October 2023, filed for bankruptcy in March 2025, and its corporate successor agreed to pay $18 million to 42 states in July 2026 to resolve claims stemming from the breach. While providing valuable genetic insights, the interpretation and clinical application of this data require careful consideration to avoid misinformation or misinterpretation that could impact patient health decisions. The potential for AI to exacerbate existing health disparities or introduce new forms of bias is a concern frequently raised by experts like Eric Topol, who emphasizes the need for robust validation across diverse patient populations to ensure equitable and safe AI implementation. Eric Topol on AI bias in healthcare

Regulatory Context: The FDA SaMD Framework and CDRH’s Mandate

The regulatory landscape, particularly the FDA SaMD Framework, provides a crucial lens through which to evaluate these failures. The FDA’s Center for Devices and Radiological Health (CDRH) is tasked with ensuring the safety and effectiveness of medical devices, including software as a medical device (SaMD). The framework emphasizes rigorous pre-market review and post-market surveillance, acknowledging the unique challenges posed by AI’s adaptive nature. Companies that have struggled often did so because they either underestimated the stringency of these requirements or failed to build their products with a deep understanding of clinical workflows and patient safety implications from the outset. Harlan Krumholz, a prominent voice in healthcare innovation, has consistently advocated for rigorous evidence generation in digital health, echoing the FDA’s emphasis on clinical validation. He stresses that for AI to be truly beneficial, it must not only be technically sound but also demonstrably improve patient outcomes without introducing new risks (DP13). The failures of companies like Forward Health, which offered a tech-forward primary care model, often stemmed from a misalignment between ambitious technological visions and the practical realities of delivering safe, effective, and reimbursable healthcare. Forward Health quietly shut down in 2026.

The Path Forward: Clinical Validation as the Cornerstone of Sustainable AI

The consistent thread running through the $62B+ graveyard of failed healthcare AI companies is a fundamental underestimation, or outright disregard, for rigorous clinical validation and patient safety. Whether it was Theranos’s fraudulent claims, Olive AI’s inability to deliver on its promises, or Pear Therapeutics’ commercial struggles despite regulatory clearances, the lesson is clear: innovation without demonstrable, clinically validated safety and efficacy is unsustainable. Investors and patient safety advocates alike must demand a higher standard. The future of responsible AI in healthcare hinges on a commitment to evidence-based development, adherence to frameworks like the FDA SaMD, and a relentless focus on improving patient outcomes safely and equitably. The next generation of successful healthcare AI companies will be those that prioritize clinical validation as their bedrock, rather than an afterthought. FDA SaMD guidance Peer-reviewed analysis of AI health failures