Healthcare AI vendors are meant to be faster, to touch more data than the systems they are replacing. The value proposition is speed and reach, and the value is tremendous. They are also why a single compromised inbox at one of these companies can put more records at risk faster than any single hospital breach.

An Xsolis breach earlier this year followed a familiar pattern; someone likely clicked a malicious link, and the threat actor’s plan was put into action. The breach, according to Paubox’s report on the incident, affected only a small part of the network environment and was quickly contained, but still affected almost 1.4 million people. The multiplier can be scaled. The vendor has hundreds of covered entities, and a single phishing email is a supply chain event.

Paubox’s 2026 Healthcare Email Security Report found that in 2025, 28% of email-related healthcare breaches originated from a vendor or business associate instead of the covered entity. AI-enabled platforms are, by design, in that vendor space, processing and transmitting PHI for the organizations that hired them.

 

The data backs up what the headlines suggest

Xsolis is part of a much larger trend that Paubox’s research has been tracking across the industry. The Healthcare Email Security Maturity Index 2026, Paubox's benchmark for U.S. healthcare IT leaders, revealed that 64% of healthcare organizations have already been the victim of an AI-generated or AI-enhanced email attack. Only 38% deployed and actively monitoring AI-based email threat detection. The result is a 26-point gap between what organizations are experiencing and what they’ve actually implemented to stop it.

The same benchmark also showed a more difficult problem where confidence and reality do not match. All the IT leaders surveyed gave Excellent or Good ratings to their real-time breach detection. In the same group, 58% had been the victims of an email breach in the last 24 months, and 23% had been breached more than once.

“We saw a specific control failure that led to this,” said Hoala Greevy, Founder & CEO of Paubox. Encryption and recipient experience were the lowest dimensions in the benchmark. When staff routes around controls that create too much friction, “Recipient experience is not secondary to security,” said Greevy.

Separately, Paubox's 2026 healthcare email security report found that 74% of breached organizations lacked effective DMARC enforcement, the control that instructs receiving mail servers how to handle messages that fail authentication checks. Without it, spoofed and AI-crafted phishing emails reach staff inboxes largely unchecked. Paubox's researchers concluded plainly that future incidents are more likely to come from environments where "the same misconfigurations and security gaps have existed for years" than from any new attack technique.

 

What the academic research adds

Peer-reviewed research backs up that the weak point is rarely the technology stack. It’s the moment that one decides to trust a message or not. A review of hospital cybersecurity published in Frontiers in Digital Health found that attackers most commonly utilize phishing as their delivery method to penetrate healthcare systems and that 89% of cybercrimes in the industry are started through phishing emails. Even organizations with strong security protocols in place still see a large percentage of successful phishing attempts.

A more sobering note on training comes from a JAMA Network Open study of phishing simulations conducted at six U.S. health care institutions. Repeated simulation campaigns were associated with reduced odds of an employee clicking a subsequent phishing email, but the effect was marginal. Even for the employees at highest risk, those who had already clicked on five or more simulated phishing emails, only 17.9% refrained from clicking any of 20 follow-up campaigns.

 

Where OCR enforcement is actually pointing

The Office for Civil Rights (OCR) has been clear about where the risk comes from. In announcing OCR’s first HIPAA penalty ever directly related to a phishing attack, then-Director Melanie Fontes Rainer said that phishing is “the most common way that hackers gain access” to health systems. In the case of Lafourche Medical Group, a Louisiana provider, a breach occurred due to phishing attacks that affected 34,862 records.

What OCR has consistently penalized in its subsequent settlements is the lack of a documented, accurate, enterprise-wide risk analysis that should have identified email as a foreseeable threat vector before the breach occurred. In April 2026, OCR resolved four different ransomware cases involving $1,165,000. In each of the four cases, the primary finding was a missing or inadequate risk analysis.

So the enforcement pattern is broader than if a specific employee clicked on a malicious link. The OCR is reviewing whether the organization identified phishing, ransomware, and email account compromise as foreseeable risks; assessed where ePHI entered, resided, and moved throughout its systems; and implemented documented measures to mitigate those risks prior to an incident.

 

The gap between confidence and control

The throughline across breach data, the benchmark research, and the peer-reviewed literature is consistent. Phishing keeps working because the organizations facing it, including sophisticated AI-driven vendors, are still relying on defenses that assume a human will consistently make the right call under pressure.

Paubox’s own research and reporting has followed this gap through dozens of incidents in 2025 and 2026. The company’s Email Suite is built around solving the inbound detection and authentication gaps. Outbound PHI HIPAA compliant email encryption. Any organization that is considering that type of tooling should start where OCR starts, with a current, honest risk assessment that names email and phishing specifically as a foreseeable danger.

 

FAQs

Why are healthcare AI vendors especially attractive phishing targets?

Healthcare AI vendors often process large volumes of PHI for multiple hospitals, health plans and other covered entities. A compromised vendor account can therefore expose information connected to many customers, turning a single phishing incident into a broader supply chain event.

 

Why are AI-generated phishing attacks harder to detect?

AI can help attackers produce polished, convincing and context-specific messages with fewer spelling mistakes or obvious warning signs.

 

Should phishing be included in a HIPAA risk analysis?

Organizations should specifically assess email-based threats, compromised credentials, domain spoofing, vendor access, account permissions and the movement of PHI through email systems.