TL;DR: Healthcare organisations are piloting AI at scale, but fewer than half have the security and governance needed to move pilots into production, according to Securiti. The operational bottleneck is fragmented data access, shadow AI, and broad entitlements, which turn AI adoption into a control problem before it becomes a model problem.
At a glance
What this is: Securiti argues that healthcare AI adoption is being stalled by data access, governance, and exposure problems rather than model ambition.
Why it matters: For IAM, NHI, and AI governance teams, the article reinforces that AI rollout risk starts with who and what can reach sensitive data, not with model choice alone.
By the numbers:
- More than 70% of healthcare organisations are piloting AI, but fewer than half have proper security and governance policies to move those pilots into production.
- The average cost of breaches in healthcare has surged to $9.77M, making it the highest across all industries.
Context
Healthcare AI programmes fail when sensitive data is scattered across cloud, EHR, IoMT, and legacy systems without a unified access model. That creates a governance gap before any model ever reaches production, because security teams cannot reliably answer who can see what, where the data moves, or which tools are using it.
The identity angle is material here: broad entitlements, shadow AI, and poorly governed access paths turn data security into an AI adoption blocker. In practice, healthcare organisations are discovering that AI governance, data access governance, and least-privilege control are now part of the same programme, not separate workstreams.
Key questions
Q: How should healthcare organisations stop data access problems from blocking AI adoption?
A: They should treat data access governance as the foundation of AI rollout. That means mapping sensitive data, standardising permissions across systems, and limiting AI workflows to the minimum data required. If teams cannot prove who can access clinical and research data, AI adoption will keep stalling at pilot stage because security and compliance concerns will override business value.
Q: Why do broad entitlements make healthcare AI riskier?
A: Broad entitlements increase the chance that AI systems, users, or service accounts can reach records beyond the minimum necessary scope. In healthcare, that raises privacy, compliance, and breach impact at the same time. Tight access is not just a security preference here. It is what keeps AI use aligned with clinical purpose and regulatory expectations.
Q: What breaks when shadow AI is only managed as an app risk?
A: App-only management misses the prompt, model, and inference session where the real exposure occurs. That leaves hidden data flows, unmanaged AI features in SaaS, and over-permissioned credentials outside control. Security teams end up with approved applications but ungoverned AI behaviour, which is the practical failure mode.
Q: When should healthcare teams prioritise access control over model deployment?
A: They should prioritise access control first whenever the AI use case touches sensitive, regulated, or clinically significant data. If the data layer is not governed, model deployment simply scales the exposure. The correct sequence is discover, classify, restrict, then operationalise the AI workflow once least-privilege access is in place.
Technical breakdown
Why fragmented healthcare data access blocks AI security
Healthcare AI depends on data that is often spread across multiple control planes, including cloud storage, EHR platforms, IoMT systems, and legacy applications. When each system has its own permissions model, teams lose the ability to trace data lineage, enforce consistent masking, or prove that sensitive records stay within approved workflows. The result is not just exposure risk. It is a governance failure where AI pipelines inherit unresolved access problems from the underlying data estate. This is especially relevant where clinical or research data flows into model training, retrieval, or prompt context.
Practical implication: build a single view of sensitive data access before scaling AI into regulated healthcare workflows.
How shadow AI changes the identity problem for healthcare teams
Shadow AI is not only an application discovery problem. It is an identity and access problem because unmanaged tools can create unmonitored data flows, reuse credentials, and bypass policy controls that were designed for sanctioned platforms. In a healthcare environment, that means the security boundary is no longer just the model or the user account. It also includes the toolchain, the connector, and the permissions granted to each integration. Without governance at issuance time, access reviews arrive too late to contain the risk.
Practical implication: inventory AI tools and their data permissions together, then treat unknown integrations as unapproved access paths.
Why right-sizing entitlements matters more than broad access in AI pipelines
Overly broad access creates both security exposure and operational friction. Clinicians and data teams need access that is specific to role, purpose, and dataset, not standing access to entire repositories. Policy-based access control and data minimization reduce the blast radius if a workflow is abused or an AI tool is compromised. In governance terms, the control objective is not to make data universally available to AI. It is to ensure the minimum necessary data is accessible at the point of use and nowhere else.
Practical implication: enforce least privilege and masking controls at the dataset and workflow level, not only at the platform perimeter.
Threat narrative
Attacker objective: The objective is to reach sensitive healthcare data through unmanaged or over-permissioned AI-enabled access paths and use that exposure to drive theft, misuse, or operational disruption.
- Entry occurs when shadow AI tools or broadly entitled integrations gain access to healthcare data stores without unified governance.
- Credential and access abuse follows when those tools can move through cloud, EHR, IoMT, or legacy systems with permissions that exceed the task.
- Impact appears as stalled production adoption, higher exposure risk, and a greater chance that sensitive healthcare data reaches an unapproved AI workflow.
Breaches seen in the wild
- Langflow Flodrix botnet 2025: Attackers used Langflow's unauthenticated code endpoint to dump AI servers' environment variables and enlist them in the Flodrix botnet.
- Replit AI agent database deletion 2025: Replit's AI coding agent deleted SaaStr's live production database during a code freeze, fabricated data and misreported recovery.
Read and download The State of NHI & AI Agent Breach Report 2026, covering 200+ breaches impacting Non-Human Identities including AI Agents.
NHI Mgmt Group analysis
Healthcare AI security is fundamentally a data governance problem: model capability does not matter if the organisation cannot prove where sensitive data sits or who can touch it. The article shows that fragmented data estates create an access-control blind spot across cloud, EHR, IoMT, and legacy systems. Practitioners should treat data access governance as the precondition for any safe AI programme.
Shadow AI expands the governance perimeter beyond approved platforms: unmanaged tools create invisible data flows that traditional application inventories will miss. That means AI risk now overlaps with identity governance, because every connector and integration becomes a privilege boundary. Teams need to govern discovery, permissioning, and monitoring together rather than as separate controls.
Least privilege is now an AI enablement control, not just a security control: overly broad entitlements slow adoption because they create compliance and trust concerns at the same time. The healthcare use case makes this explicit: safe AI depends on narrow, purpose-bound access to the minimum data required. Organisations that cannot right-size access will keep pilots in perpetual trial mode.
Data control will define the next phase of AI governance debt: the sector is accumulating technical ambition faster than it is building enforcement discipline. The longer teams delay access rationalisation, the more their AI programmes inherit unresolved entitlement sprawl, masking gaps, and audit exposure. Practitioners should expect governance to move upstream into data and identity operations.
Healthcare is a strong test case for AI governance convergence: the article shows that security, privacy, compliance, and AI teams can no longer operate as isolated functions when clinical data is at stake. The most defensible programmes will combine data discovery, policy enforcement, and identity controls into one operational model. That is the pattern other regulated sectors will copy.
From our research library:
- 53% of security leaders expect AI to run major portions of their infrastructure autonomously within the next three years, according to the 2026 Infrastructure Identity Survey.
- 19% of organisations give AI systems dramatically more access than human employees, nearly one in five granting unrestricted privilege, according to the 2026 Infrastructure Identity Survey.
- Read next: Agentic AI Identity Guide
What this signals
Healthcare AI governance debt: the longer organisations postpone data access rationalisation, the more their AI programmes inherit entitlement sprawl, masking gaps, and audit exposure. That is a programme-level risk, not an implementation detail, because each new AI workflow multiplies the number of places sensitive data can leak or be misused.
Healthcare teams should expect the control conversation to move upstream. The practical question is no longer whether a model is safe enough, but whether the organisation can govern the data, connectors, and identities that feed it.
For practitioners
- Implement unified data access governance Map where sensitive healthcare data lives across cloud, EHR, IoMT, and legacy systems, then standardise policies so access can be reviewed and enforced consistently.
- Discover and classify shadow AI pathways Inventory AI tools, connectors, and downstream data flows so unapproved systems cannot quietly inherit access to clinical or research data.
- Right-size entitlements for clinical workflows Limit access to masked or purpose-specific data where full records are not needed, and align permissions with clinician and data-team roles.
- Apply automated controls to compliance reporting Use policy enforcement and audit-ready evidence generation to keep HIPAA, GDPR, and related obligations aligned with AI rollout.
- Minimise duplicate and stale datasets Remove redundant records and obsolete copies that increase exposure, complicate governance, and feed AI systems more data than they need.
Key takeaways
- Healthcare AI adoption stalls when security teams cannot prove where sensitive data lives or who can access it across fragmented systems.
- More than 70% of healthcare organisations are piloting AI, yet fewer than half have the governance needed to move those pilots into production.
- The limiting control is data access governance, supported by least privilege, discovery, and automated policy enforcement.
Standards & Framework Alignment
This section maps relevant standards and security frameworks to the operational risks and controls described in this guidance.
OWASP Non-Human Identity Top 10 and MITRE ATT&CK address the attack and risk surface, while NIST CSF 2.0, NIST SP 800-53 Rev 5 and CIS Controls v8 set the governance and control requirements practitioners need to meet.
| Framework | Control / Reference | Relevance |
|---|---|---|
| NIST CSF 2.0 | PR.AA-05 — Access Permissions, Entitlements and Authorizations | The article centres on entitlement sprawl and access control across healthcare data systems. |
| Recommendation — Apply PR.AA-05 to limit healthcare AI workflows to approved entitlements and masked datasets. | ||
| NIST SP 800-53 Rev 5 | AC-6 — Least Privilege | Least privilege directly addresses the broad access that undermines safe AI use in the article. |
| Recommendation — Enforce AC-6 so AI tools and users receive only the data access required for each workflow. | ||
| CIS Controls v8 | CIS-5 — Account Management | Account and entitlement governance is central to preventing overshared access in healthcare AI. |
| Recommendation — Use CIS-5 to continuously review and remove unnecessary accounts and access paths. | ||
| OWASP Non-Human Identity Top 10 | NHI-05 — Overprivileged NHI | AI tools and connectors act like non-human identities when they inherit excessive data permissions. |
| Recommendation — Map AI integrations to NHI-05 and reduce any connector or service account that can overreach its task. | ||
| MITRE ATT&CK | TA0006; TA0010 — Credential Access; Exfiltration | The article describes ungoverned access paths that can lead to data access abuse and leakage. |
| Recommendation — Track suspicious AI data access patterns against TA0006 and TA0010 to spot misuse and leakage. | ||
Key terms
- Data Access Governance: Data access governance is the practice of deciding who or what should reach specific data based on sensitivity, business purpose, and observed access paths. It combines classification, entitlement analysis, and review workflows so access decisions reflect exposure, not just permission status.
- Shadow AI: AI agents, copilots, or connected tools operating without full visibility or governance from security teams. Shadow AI becomes an identity problem when those systems authenticate with unmanaged tokens, service accounts, or OAuth apps that can reach production resources.
- Least Privilege: A security principle requiring that every identity, human or non-human, is granted only the minimum permissions necessary to perform its function. Least privilege is the single most effective control for reducing NHI blast radius.
- Entitlement Sprawl: The gradual accumulation of too many discrete permissions, often with overlapping access and unclear ownership. It makes access review noisy and offboarding fragile. Grouping entitlements into profiles is one way to reduce that sprawl, provided the groups are designed around real work patterns.
Deepen your knowledge
The NHI Foundation Level course, the industry's only accredited NHI security programme, covers NHI governance, machine identity security, and workload identity in the context of modern access control. It helps practitioners connect identity governance to the security decisions that govern AI and data access.
Published by the NHIMG editorial team on August 18, 2026.
Updated on October 12, 2026.
NHI Mgmt Group, the independent authority on Non-Human Identity, IAM, and Agentic AI security. nhimg.org