Healthcare firms fail to monitor AI tools, even as shadow AI begins leading to breaches

Cybersecurity Dive

Source published: Site published:

Original article

Published by site logo
News Brief

The healthcare industry is adopting new technology, but security teams have too much work to monitor all of it.

Published Nov. 17, 2026
Eric Geller Senior Reporter
Glowing blue AI microchip inside a digital shield, surrounded by icons representing authentication, encryption and network security.
Getty Images

News Brief:

  • Healthcare companies face serious cybersecurity risks because they use AI systems without carefully watching them, security company DNSFilter said in a report published Tuesday.
  • Healthcare providers understand some of these risks, such as employees using AI tools in unsafe ways. However, they are not doing enough to make sure these tools are used safely and work as they should.
  • The report shows how new technology and long-standing cybersecurity problems are coming together to threaten the ability of healthcare organizations in the U.S. to provide reliable care.

What It Means:

Healthcare is one of the main targets for hackers. Hospitals and other providers store valuable information and cannot easily stop working. Also, healthcare organizations depend on many outside companies and services. These large supply chains give attackers many ways to attack. Experts have repeatedly warned healthcare organizations to make supply-chain security a top priority. But many still do not actively check the security risks created by outside companies they work with.

Six out of 10 people who answered DNSFilter's survey said their organizations depended on their AI and software-as-a-service (SaaS) providers to tell them when a security problem happened. This was much higher than the average of 43% among other industries DNSFilter had surveyed. Fewer than one in four respondents (23%) said their organizations depended on employees to report security problems involving AI systems and SaaS platforms.

DNSFilter's report is based on answers from 30 IT and cybersecurity professionals working in healthcare. They took part in a survey in June 2026. The survey reached 400 managers across several industries in total.

Healthcare companies often have too few people working in security. These workers often have too much to do, making it hard to watch systems that look for hackers and other threats. In DNSFilter's survey, only 27% of respondents said they were sure their organization watched all its security tools. Four out of 10 said they watched fewer than half of their tools. In DNSFilter's surveys of other industries, only 23% said they watched fewer than half of their tools.

DNSFilter warned: “A tool nobody is watching can't raise an alert.”

Shadow AI is a serious problem in important infrastructure industries, and healthcare is no exception. Shadow AI means AI tools or connections that employees use without their organization's permission or knowledge. Twenty-seven percent of IT and cybersecurity professionals at healthcare organizations said they knew about AI connections to their systems that had not been approved. Another 23% thought this had happened but could not prove it.

Making sure employees follow company rules is an important way to stop shadow AI from spreading. But although 70% of healthcare IT and cybersecurity managers said their organizations had rules for using AI, only 40% said their companies always enforced those rules. DNSFilter said this was the lowest rate of any industry.

At almost half (47%) of healthcare organizations, employees can approve a new AI tool's access to company systems without getting permission from senior leaders. DNSFilter called this “especially concerning in an industry where the critical data needing protecting is protected health information.” One in 10 people who answered the survey said an AI tool that had not been approved had already caused a cybersecurity incident.

Healthcare IT managers said they were confident they could find security incidents quickly. Eighty-seven percent said they believed they could discover a data breach within six hours. This included 40% who said they were very confident. However, one-third of respondents also said they had experienced a security incident in the past 12 months that started when an AI or SaaS tool was compromised. Another one-third said they had experienced an incident that they never reported to regulators or customers.