Overview
The Healthcare and Public Health Sector Coordinating Council (HSCC) has published a new guidance document addressing cybersecurity risks that arise specifically from artificial intelligence deployments in healthcare settings. The guide reflects growing concern among sector leaders that existing regulatory frameworks — including HIPAA's Security Rule — were not designed with AI-specific threat vectors in mind and leave provider organizations without adequate direction for governing AI tools responsibly.
HSCC leaders emphasized that the pace of AI adoption across clinical decision support, administrative automation, and patient-facing applications has outrun the governance structures most provider organizations have in place. The new guide is intended to fill that gap with practical, operationally focused direction rather than broad compliance principles.
The publication arrives as federal regulators and standard-setting bodies are still developing formal AI-specific requirements for healthcare. Until those frameworks mature, HSCC argues that provider organizations must build their own internal governance structures — and the guide offers a framework for doing so.
Key developments
Regulatory guidance has not kept pace with deployment. HSCC's core argument is that HIPAA and existing HHS guidance address data privacy and security at a structural level but do not account for risks introduced by AI systems, such as model poisoning, training-data exposure, or opaque algorithmic outputs affecting clinical decisions.
The guide targets governance, not just technical controls. Rather than cataloguing known AI attack techniques, HSCC frames the document around governance — how provider organizations should assign accountability, evaluate AI vendors, monitor deployed models, and respond when AI behavior deviates from expected parameters.
Vendor relationships receive explicit attention. Because most AI tools in healthcare are procured from third-party vendors, the guide addresses the business associate and procurement dimensions of AI risk. Provider organizations are urged to scrutinize vendor contracts, audit rights, and data-handling practices for AI systems specifically, not just for traditional software.
The guide applies across organization sizes. HSCC's framing does not assume enterprise-level IT infrastructure, which means independent practices procuring AI-enabled EHR features, clinical documentation tools, or patient communication platforms fall within the intended audience.
## Industry impact
AI adoption in healthcare is accelerating across settings of all sizes. Clinical AI scribes, automated prior-authorization tools, and AI-assisted diagnostic imaging are increasingly available to small and mid-sized practices through EHR integrations and third-party modules. That accessibility also means that practices with limited IT staff are now operating AI systems whose underlying data flows and risk profiles they may not fully understand.
The HSCC guide reflects a sector-wide recognition that standard cybersecurity frameworks — NIST CSF, HIPAA's Security Rule safeguards, and SOC 2 certifications — do not map cleanly onto AI-specific risks. Model integrity, training-data provenance, and inference-time manipulation are distinct threat categories that existing checklists do not address. The HHS Office for Civil Rights has not yet issued AI-specific HIPAA guidance; until it does, HSCC documents represent the most operationally grounded direction available to the sector.
Independent practices that have adopted AI tools through vendor bundles may have done so without recognizing that those tools introduce new data-sharing arrangements and potential PHI exposures that warrant explicit attention in their risk analysis.
What this means for independent practices
- Inventory AI tools currently in use. This includes AI features embedded in EHR platforms, patient messaging tools, coding and billing automation, and any third-party clinical decision support modules. Many practices have enabled AI features without conducting a separate risk assessment for those capabilities. - Update business associate agreements to address AI. If a vendor processes PHI through an AI system, the BAA should specify how training data is handled, whether PHI is used to improve the model, and what audit rights the practice retains.
- Assign internal accountability for AI governance. Someone within the practice — or a contracted compliance officer — should be responsible for monitoring AI tool performance, reviewing vendor communications about model updates, and documenting how AI outputs are reviewed before clinical or administrative action is taken.
- Include AI systems in the annual HIPAA risk analysis. AI tools that access, process, or transmit PHI are subject to the same Security Rule requirements as any other system. The risk analysis should identify AI-specific threat scenarios, not just traditional breach pathways.
- Scrutinize vendor contracts before enabling new AI features. Before activating AI-driven features offered through existing vendor relationships, practices should verify data-use terms, confirm whether PHI is retained for model training, and ensure they can disable the feature without losing access to the underlying system.
Practices that have treated AI features as passive software add-ons rather than distinct data-processing arrangements are carrying unexamined exposure. Effective AI governance is an ongoing operational discipline — it requires regular review of what AI tools are active, what data they touch, and whether the vendor's handling of that data has changed since the tool was first adopted.
What would have prevented this
AI-specific risk assessments: Standard HIPAA risk analyses address data storage, transmission, and access controls. A dedicated AI risk assessment layer examines training-data exposure, model update procedures, inference-time data flows, and the clinical consequences of erroneous AI outputs.
Contractual audit and data-use controls: Business associate agreements and vendor contracts for AI tools should specify whether PHI is retained beyond the immediate transaction, who owns model improvements derived from the practice's data, and whether the practice has audit rights over the vendor's AI processing environment.
Role-based access controls (RBAC) applied to AI outputs: Clinical and administrative staff should access AI-generated outputs only within their scope of work. AI systems that surface patient data as part of their output should be subject to the same access-control requirements as any other PHI-containing system.
Model monitoring and anomaly detection: Deployed AI tools should be subject to ongoing performance monitoring so that significant deviations in output — which may indicate data-integrity problems or adversarial interference — are flagged before they affect clinical or billing decisions.
Vendor due diligence specific to AI procurement: Before adopting any AI tool that touches PHI, practices should evaluate the vendor's data governance documentation, security certifications, incident history, and procedures for notifying customers of model changes that affect data handling.