After two U.S. hospital systems allegedly lost nearly $7.8 million in an executive impersonation scheme, healthcare cybersecurity leaders say increasingly convincing AI-generated voices and videos are pushing hospitals to rethink how employees verify requests from executives.
The U.S. Attorney’s Office for the Southern District of New York announced Nov. 5 that an indictment had been unsealed charging three foreign nationals in an international fraud scheme that allegedly used advanced technology to impersonate corporate executives and attorneys. The scheme allegedly resulted in more than $129 million in losses worldwide, including approximately $4.83 million from a Missouri-based hospital system and $2.97 million from a California-based system.
Prosecutors alleged the defendants and their co-conspirators posed as CEOs who needed help with confidential financial transactions, using technology described as “super intelligence” to mimic the executives’ voices or images.
The case highlights a growing challenge for healthcare organizations as AI makes traditional warning signs of impersonation, including unfamiliar voices, unusual language and suspicious video, less reliable.
Six hospital and health system cybersecurity leaders told Becker’s that their response increasingly centers on independent verification, established approval procedures and empowering employees to question requests from senior executives.
From spotting fakes to verifying requests
Chris Stucker, deputy chief information security officer at Milwaukee-based Froedtert ThedaCare, said AI-generated content has made many of the warning signs traditionally emphasized in cybersecurity training less dependable.
“Rather than expecting team members to become human lie detectors, we are shifting our focus from perception to process,” Mr. Stucker told Becker’s. “Even if someone can spot a deepfake today, that may not be possible tomorrow as the technology continues to improve.”
At Froedtert ThedaCare, employees are trained to follow established verification procedures for requests involving money, credentials, payroll, access and other sensitive matters. That includes contacting individuals through known phone numbers or approved communication channels rather than relying on information provided in the original request.
Mr. Stucker said that emphasis on caution also carries a risk in healthcare settings. Employees might hesitate to respond to legitimate urgent requests from physicians or executives out of concern that the communications are fraudulent.
Eric Sean Clay, vice president and chief security officer at Baton Rouge, La.-based FMOL Health, described a similar shift. He said cybersecurity awareness programs have traditionally focused on identifying suspicious emails and other signs of fraud, but AI is making impersonation attempts more convincing.
As a result, education must extend beyond IT and finance to clinical leaders, executive assistants, human resources, revenue cycle teams and help desks, all of which may receive requests involving sensitive information or financial transactions, Mr. Clay said.
“We’re moving from spot the fake to verify the request,” Mr. Clay told Becker’s. “With AI, technology will keep improving, so we can’t build a defense on teaching people to hear glitches.”
Mr. Clay emphasized independently confirming suspicious requests and requiring multiple approvals or established authorization procedures for high-risk actions, including wire transfers, banking changes, password resets and access to patient records.
He also said executives must reinforce employees’ authority to question requests, regardless of seniority.
“I also have a message for executives: Tell your teams you expect to be verified, and never get irritated when it happens. If leaders bypass their own controls, the controls fail,” he said.
Existing controls take on new importance
At Louisville, Ky.-based UofL Health, AI-enabled impersonation is not fundamentally changing the organization’s cybersecurity awareness strategy, but it is shifting training priorities, according to Matt Phillips, system director and chief information security officer.
UofL Health already provides training on phishing and voice phishing, also known as vishing. The organization is adjusting its messaging to account for AI-generated threats while reinforcing existing safeguards, including segregation of duties and approval procedures.
“We never do anything from a business perspective because one leader said it,” Mr. Phillips told Becker’s.
He said requests must move through established processes, such as creating a ticket and obtaining the necessary approvals, rather than being carried out solely on an executive’s instructions. Those procedures apply even when a request appears urgent or comes from someone claiming to hold a senior leadership position, he said.
Mr. Phillips said the same emphasis on verification extends to hiring, where organizations are encountering situations in which the person represented on a resume is not the person who appears for an interview.
UofL Health is also evaluating technologies capable of detecting AI-generated voices and videos, but Mr. Phillips said those tools have limitations and can be costly. For now, the organization is prioritizing existing procedures and controls that validate transactions and identify potential fraud.
Hospitals put AI impersonation into employee training
At Topeka, Kan.-based Stormont Vail Health, cybersecurity training is evolving to address AI impersonation more directly.
Joe Susai, chief information security officer, said the organization regularly updates its learning management system training, discusses cybersecurity during employee orientation, adjusts phishing simulations and distributes security bulletins.
During Cybersecurity Awareness Month, Stormont Vail dedicated an entire week to AI impersonation and spoofing, including warning signs and appropriate employee responses.
“The message is simple: Verify the request, not just the person,” Mr. Susai told Becker’s.
Employees are encouraged to independently confirm requests involving money, protected health information, credentials or other sensitive information, particularly when someone pressures them to bypass normal procedures.
At Providence, R.I.-based Care New England, Brian Schilling, chief information security officer, said the organization is similarly emphasizing verification and employees’ ability to challenge unusual requests.
He pointed to a 2024 incident involving engineering firm, in which an employee in Hong Kong was deceived by a fabricated video conference featuring individuals impersonating company executives and colleagues. The incident resulted in approximately $25 million in fraudulent transfers.
For hospitals, Mr. Schilling said, the potential consequences extend beyond financial losses. Executive impersonation could expose patient information, provide unauthorized access to clinical systems or disrupt operations.
“Ultimately, we can’t expect every employee to identify a convincing AI-generated threat,” Mr. Schilling told Becker’s. “What we can do is make sure they know when to stop, question a request and verify it before taking action.”
He said organizations must also establish safeguards so a single fraudulent communication cannot lead directly to a financial transfer, disclosure of sensitive information or changes to critical systems.
The next concern is scale
Beyond increasingly convincing impersonations, cybersecurity leaders are concerned about the growing volume and accessibility of AI-enabled attacks.
Mr. Phillips said UofL Health has encountered suspected incidents that employees were able to identify and report, allowing the organization to adjust its training and communications. His biggest concern is the possibility that these attempts could become significantly more widespread.
Mr. Stucker raised a related concern about employee fatigue. As attacks become faster, less expensive and more convincing, he said, employees may become overwhelmed by repeated threats.
Matthew Snyder, chief information security officer at Hershey, Pa.-based Penn State Health, said the underlying problem is broader than any single deepfake technology. Attackers can combine AI with publicly available information and established social engineering tactics to create highly personalized requests that appear credible.
Organizations cannot expect employees to become experts at identifying synthetic media, he said. Instead, security programs must account for the possibility that a fraudulent request may appear entirely authentic.
“Awareness has to evolve with the threat. We need to teach people not only how to spot something suspicious, but also what to do when a request looks completely legitimate and still may not be,” Mr. Snyder told Becker’s.
For Mr. Clay, the challenge also extends beyond executive impersonation. Physicians, nurse leaders, vendors and even patients’ family members could be impersonated, potentially creating risks involving patient information, clinical operations and public trust.
That makes verification more than a cybersecurity training exercise, he said. It requires coordination across technology, compliance, security and clinical leadership.
“The bottom line is that our workforce must move from ‘trust, then verify’ to ‘verify, then trust,’” Mr. Clay said.