Healthcare organizations struggle to monitor AI and SaaS security risks

src-b1358f33

Healthcare organizations are adopting artificial intelligence tools while many security teams lack consistent visibility into how those tools and connected software-as-a-service platforms are used. A DNSFilter report based on a survey of healthcare IT and cybersecurity professionals describes gaps in monitoring, policy enforcement and incident reporting that could leave sensitive information exposed.

The findings highlight a particular concern for data protection: employees may connect unapproved AI services to organizational platforms, while security staff may not have the capacity or processes to identify those connections. In a sector responsible for protected health information and dependent on technology to deliver care, weak oversight can complicate efforts to prevent data loss and respond to incidents.

Visibility into tools remains limited

Only 27% of healthcare respondents said they were confident their organizations monitored all of their security tools. Four in 10 said they monitored fewer than half. That share was higher than the 23% reported in DNSFilter research involving other industries. The report cautioned that tools without active oversight cannot reliably generate useful alerts for responders.

Staffing constraints are part of the picture, according to the report: healthcare security personnel are often stretched thin, making it difficult to monitor intrusion-detection systems and other controls closely. The survey results do not identify which specific tools were left unmonitored, or establish that a particular monitoring gap caused an incident. They do, however, indicate that many organizations may lack a complete view of their security environment.

Reliance on vendors to disclose incidents is another potential weak point. Six in 10 healthcare respondents said they depended on AI or SaaS vendors to notify them about security incidents, compared with 43% in DNSFilter’s surveys of other sectors. Only 23% said they relied on employees to report incidents involving those services.

Vendor notifications are an important part of incident awareness, but the survey figures show that many organizations expect providers to initiate communication. Security teams should consider how they would detect or validate a problem if a vendor notification is delayed or incomplete, and whether employees know where and how to report suspicious activity. The report does not provide details about respondents’ contractual requirements or incident-notification processes.

Shadow AI policies are not consistently enforced

Unauthorized AI connections were a known or suspected issue for many respondents. Twenty-seven percent said they were aware of unauthorized AI connections to their organizations’ platforms, while another 23% suspected such connections had occurred but could not confirm them.

Although 70% said their organizations had AI policies, only 40% said those policies were enforced consistently. DNSFilter said this was the lowest level of consistent enforcement among the sectors it surveyed. In nearly half of the healthcare organizations represented, 47%, employees could authorize a new AI tool to access company systems without higher-level approval.

Policy documents alone do not establish which services employees are using or what information they can share with them. The survey points to a need for organizations to connect policy with practical controls and oversight: define who can approve AI tools, review access to organizational systems, and make reporting routes clear to staff. These are defensive considerations prompted by the reported gaps, not measures the survey said every organization had implemented.

One in 10 respondents said an unapproved AI tool had already led to a cybersecurity incident. Separately, one-third reported an incident in the previous 12 months that began with a SaaS or AI tool compromise. The report does not specify the nature of those incidents, whether they involved data exposure, or whether the incidents overlapped with the unapproved-tool cases. The figures should therefore be read as indicators of reported experience, not as a measure of industry-wide breach prevalence.

Detection confidence does not resolve reporting gaps

Healthcare respondents expressed confidence in their ability to identify incidents quickly: 87% said they were confident they could detect a breach within six hours, including 40% who were very confident. At the same time, one-third said they had experienced an incident in the past year that they never reported to regulators or customers.

The survey does not explain why those incidents went unreported or whether reporting was legally required in each case. Still, the contrast underscores that detection speed is only one part of incident readiness. Organizations also need clear processes to assess an event, determine applicable notification obligations and document decisions.

DNSFilter’s findings come from 30 IT and cybersecurity professionals in healthcare who answered a June 2026 survey that reached 400 managers across multiple industries. The small healthcare respondent group provides a snapshot rather than a comprehensive measure of all providers. For security leaders, its central message is operational: track AI and SaaS access, enforce approval policies consistently, and avoid relying on a single source—whether employees or vendors—to reveal security incidents.