Incident sheet · /security

Filed
Read
4 min · 838 words
Severity
Routine

Healthcare AI governance lags behind rising patient-data risk

A DNS Filter survey finds 43% of healthcare leaders see employee sharing of sensitive data with AI as the top emerging risk, while only 33% will fund AI governance.

Healthcare Employees Sharing Patient Data with AI Tools Viewed as Major Emerging Risk
FeedHealthcare Employees Sharing Patient Data with AI Tools Viewed as Major Emerging Risk — fernandozhiminaicela / pixabay

Event log

  1. 43% of healthcare security leaders named employee sharing of sensitive data with AI models as the biggest emerging risk over the next 12 months.
  2. 27% confirmed and 23% suspected unapproved AI connections to Google Workspace, Microsoft 365 or single sign-on environments.
  3. 70% had formal AI-tool policies, but only 40% said those policies were consistently enforced.
  4. One-third had a SaaS or AI-originated incident in the past 12 months; one-third also reported an incident that was never formally reported or disclosed.
  5. Only 33% planned to put their next security dollar into AI governance, compared with 41% in other industries.

Only 33% of healthcare organizations will send their next security dollar to AI governance, even as 43% of security leaders call employee sharing of sensitive data with public AI models the largest emerging AI risk.

That mismatch defines the control problem. DNS Filter's report, Seeing the Risk, Missing the Signal: AI Security in Healthcare, examines risk recognition, policy, monitoring, detection and spending. It does not disclose its sample size, fieldwork dates or a count of incidents.

What is the biggest emerging risk?

Forty-three percent of healthcare security leaders identified employee sharing of sensitive data with AI models as the biggest emerging risk over the next 12 months. Loss of intellectual property through AI training data ranked first for 53% of healthcare respondents, compared with 31% across other sectors.

Public models such as ChatGPT, Gemini and Claude create one exposure route, while unauthorized browser extensions, plug-ins and OAuth grants create another after approval in a few clicks. Both routes can expose protected health information outside approved controls. The cited employee uploads violate organizational privacy and security policies and the HIPAA Rules.

Ken Carnesi, CEO of DNS Filter, said: “Healthcare security leaders know exactly where AI puts their data at risk. What they’re missing is the moment that risk becomes an incident, and too many are hearing about it from a vendor instead of their own team.”

“When the data involved is protected health information, the risk only becomes more significant,” he said.

Which privacy incidents are already public?

Healthcare providers have reported employee uploads of sensitive patient data to public AI models. Cited organizations include Johns Hopkins Bayview Medical Center, Weill Cornell Medicine and the Washington Department of Social and Health Services.

The HHS Office for Civil Rights breach portal publishes cases affecting 500 or more individuals, while media notices apply only at that federal disclosure threshold for covered breaches. Smaller incidents can therefore escape public view. The survey supports concern about wider exposure, but it does not quantify how many cases occurred.

How widely are staff connecting unapproved tools?

The connections involved Google Workspace, Microsoft 365 and single sign-on environments.

  • 27% confirmed that employees connected an unapproved AI tool to one of those environments.
  • 23% suspected the same connection.

Patient data may not have been involved in those confirmed or suspected cases. That caveat reduces the proven breach rate, but it does not remove the exposure created by missing AI governance. Without AI governance, privacy violations may go unnoticed.

Unauthorized browser extensions, plug-ins and OAuth grants can widen that exposure without requiring a staff member to upload a file. Forty-seven percent said an employer could authorize a new AI tool to access company systems without higher-level approval.

Do policies close the gap?

The survey findings expose a consistent gap between written controls and operational enforcement:

  • 70% of organizations had formal policies prohibiting unapproved AI tools.
  • 40% said those policies were consistently enforced.
  • 27% actively monitored every security tool they owned.
  • 40% monitored 50% or fewer security tools.

Half of respondents called AI security and governance one of their largest security gaps. That was the highest share of any sector surveyed. The policy rate looks reassuring in isolation, but the enforcement and monitoring figures describe controls operating more often on paper than in practice across surveyed healthcare organizations.

Formal approval is common. Consistent verification is less common.

Can healthcare detect and disclose AI incidents?

Six in 10 healthcare organizations relied on AI and SaaS vendors to alert them to incidents, compared with 43% in other sectors. Only 23% caught those incidents through employee reports, against 44% elsewhere, a 21-point gap that leaves healthcare more dependent on outside notification and less reliant on internal reporting.

Despite that dependency, 87% of healthcare respondents were confident they could detect a breach within six hours. Forty percent felt very confident. One-third had experienced a SaaS or AI-originated incident in the previous 12 months. The same share had experienced one that no formal process reported or disclosed.

That contradiction deserves attention: fast detection confidence sits beside substantial failure to report or disclose, though the survey gives no incident totals.

Is security funding keeping pace?

Only 33% of healthcare respondents would direct their next security dollar to AI security and governance, versus 41% in other industries. Concern has outpaced commitment. The report leaves the practical conclusion straightforward: governance remains a budget line, not yet a consistently enforced control.

Carnesi added: “Before the next dollar goes to AI governance, healthcare organizations need to see which AI services their people are already using without clearance.”

Organizations should first inventory unapproved AI services, as Carnesi recommends, although the survey cannot establish how many patient records were exposed or how often disclosure failed. No product versions, named models tied to individual incidents, sample size, fieldwork dates or breach totals appear in the account. That leaves the immediate risk credible but its scale uncertain: provider cases confirm exposure paths, while the survey mainly measures perceptions, policies and monitoring practices.

via hipaajournal.com (Source)

Filed under

  • ai-governance
  • healthcare-security
  • patient-data
  • ai-security
  • data-privacy
Share this article:

More from Daniel Okafor

Daniel Okafor

Show full bio

Staff writer covering technology and product news in enterprise information technology and the leaders who run at CIO Monitor.

201 articles

The newsroom digest

One email a day with the stories that matter. No noise.

By subscribing you agree to the privacy policy.

Cross-references

‹ Previous articleNext article ›