Beyond Clearance: Why Privacy Enforcement De-Risks Your AI Investment

The pathway to FDA clearance for an AI-powered medical device is a well-trodden, if complex, road. Yet, achieving regulatory approval for the device itself does not automatically secure the integrity of the data flowing through it, nor does it insulate vendors from privacy enforcement. This distinction, often overlooked, is becoming a critical vulnerability for health AI companies.

The Dual Imperative: Device Review and Data Stewardship

Device review, particularly for SaMD, focuses on the safety and effectiveness of the software’s intended use. This involves rigorous testing, clinical validation, and adherence to quality management systems (QMS / ISO 13485) FDA guidance on SaMD premarket submissions. The FDA’s remit is primarily product-centric: does the AI model perform as claimed, and is it safe for patients? Privacy review, conversely, interrogates the entire data lifecycle surrounding that product. It asks what data is collected, how it is stored, who has access to it, and what safeguards are in place to prevent unauthorized disclosure or misuse. The HIPAA Security Rule, enforced by the HHS Office for Civil Rights (OCR), sets the standard for protecting electronic protected health information (ePHI). This rule outlines administrative, physical, and technical safeguards that covered entities and their business associates must implement. The questions posed by a privacy officer are distinct from those of a regulatory affairs specialist, even if both in the end pertain to the same AI health tool. One asks what the model does. The other asks what happens to the data around it. A complete file, one that withstands scrutiny from both the FDA and HHS OCR, answers both.

HIPAA Security Rule: The Unseen Regulator

The HIPAA Security Rule mandates a complete approach to ePHI protection. It requires covered entities and business associates to:

• Ensure the confidentiality, integrity, and availability of all ePHI they create, receive, maintain, or transmit.
• Protect against any reasonably anticipated threats or hazards to the security or integrity of ePHI.
• Protect against any reasonably anticipated impermissible uses or disclosures of ePHI.
• Ensure compliance by their workforce. These requirements are not prescriptive regarding specific technologies but rather outcomes, though proposed updates to the HIPAA Security Rule for 2026, with a final rule targeted for July 2027, aim to make certain safeguards like encryption, multi-factor authentication, and vulnerability scanning mandatory. For an AI health tool, this translates into a need for strong technical controls (e.g., encryption, access controls, audit logs), clear administrative policies (e.g., risk analysis and management, workforce training), and physical safeguards (e.g., facility access controls). The enforcement history of HHS OCR demonstrates a consistent focus on these foundational elements, often penalizing entities for systemic failures in their security posture, rather than isolated incidents. HHS OCR has also restructured its enforcement efforts, establishing a dedicated unit for privacy and security enforcement as of May 18, 2026.
  

  The Recorded Set: HeartFlow, Tempus AI, and Butterfly Network

  The privacy enforcement field offers instructive examples of how these principles are applied. HeartFlow, Tempus AI, and Butterfly Network, while distinct in their core offerings, share a common thread in the public record: their interactions with data privacy expectations and, in some cases, the consequences of falling short. HeartFlow, for instance, develops AI-powered analysis of coronary CT angiograms to create 3D models of coronary arteries, aiding in the diagnosis of coronary artery disease. This process involves highly sensitive patient imaging data. While their product, including their Next Gen HeartFlow Plaque Analysis algorithm, has received FDA 510(k) clearance (most recently on September 22, 2025), the underlying data handling infrastructure is subject to the same stringent HIPAA requirements as any other health data processor. The focus here shifts from the accuracy of their CT-FFR analysis to the security of the imaging data itself, how it is transmitted, stored, and accessed by their AI algorithms and human operators. Any vulnerability in this data pipeline, regardless of the AI’s diagnostic precision, could trigger an HHS OCR inquiry. Tempus AI, a company focused on precision medicine through genomic sequencing and real-world data analysis, operates at the intersection of vast and complex datasets. Their AI tools analyze patient molecular and clinical data to assist oncologists in treatment decisions. The sheer volume and sensitivity of the data they handle, genomic sequences, electronic health records, and clinical trial data, places an enormous burden on their privacy and security frameworks. The company is currently facing multiple class-action lawsuits, consolidated as of April 15, 2026, alleging improper use and sharing of genetic data obtained through an acquisition, without patient consent. These lawsuits highlight the challenges of de-identification and the secure aggregation of such diverse data types, arguing that genetic information may be inherently identifiable. The question is not whether their AI can identify relevant mutations, but whether the process of acquiring, storing, and processing that data maintains patient privacy and security at every step. Butterfly Network, known for its portable ultrasound device and integrated AI, provides another lens. Their device integrates AI to assist with image acquisition and interpretation. On March 30, 2026, Butterfly Network received FDA clearance for a fully automated Gestational Age (GA) Tool integrated into its handheld ultrasound platform. The data generated, ultrasound images and associated patient information, is transmitted and stored, often in cloud environments. For a device that aims to democratize ultrasound access, the privacy implications extend to diverse clinical settings and potentially less controlled environments. The company’s success in device clearance is separate from its ongoing responsibility to ensure that the data collected via its platform is secured according to HIPAA standards, including strong encryption, secure cloud infrastructure, and clear data use agreements with its customers. These cases, taken together, illustrate a consistent theme: the FDA’s review of a device’s safety and efficacy does not substitute for, nor does it automatically guarantee compliance with, health data privacy regulations. Each entity, despite its innovative AI applications, must maintain a separate, strong privacy file that demonstrates adherence to the HIPAA Security Rule.

  The Questions a Privacy Officer Must Ask

  For privacy officers and counsel at health AI vendors, the implications are clear. The product development lifecycle must integrate privacy-by-design principles from inception, not as an afterthought to FDA clearance. This means asking critical questions that go beyond the device’s functionality:

• Data Minimization: Is only the absolutely necessary patient data being collected for the AI’s intended purpose? Can the AI function effectively with de-identified or anonymized data where appropriate?
• Data Security Architecture: What are the end-to-end security measures for all ePHI, from acquisition through storage, processing, and disposal? This includes encryption at rest and in transit, strong access controls, and regular vulnerability assessments and penetration testing. NIST cybersecurity framework for healthcare
• Vendor Management: If third-party vendors are involved in data processing or storage, are Business Associate Agreements (BAAs) in place, and do these agreements adequately transfer HIPAA obligations and liability?
• Incident Response: Is there a well-defined and regularly tested incident response plan for data breaches involving ePHI?
• Transparency and Consent: How is patient consent obtained for data use, particularly for novel AI applications, and how transparent are the data practices to both patients and healthcare providers?
• Audit Trails: Are complete audit logs maintained for all access to and modifications of ePHI, enabling forensic analysis in case of a breach?
• Training and Policies: Is the workforce adequately trained on HIPAA regulations and the company’s specific privacy and security policies? Are these policies regularly reviewed and updated? These are not questions that the FDA device review process is designed to answer in depth. They fall squarely within the domain of privacy and security compliance. Companies that fail to maintain a distinct and strong privacy file, separate from their device clearance documentation, expose themselves to significant enforcement risk from HHS OCR, irrespective of their FDA status. The path to market for an AI health tool is thus a dual one, requiring both a validated device and an unimpeachable data stewardship framework. The absence of the latter can negate the achievements of the former.