In late June, Tennessee-based Xsolis, which delivers its AI-powered Dragonfly platform for utilization management to over 600 health systems and payers, disclosed a major data breach affecting nearly 1.4 million patients at associated hospitals. Through a planned and targeted phishing operation, threat actors gained access to the names, addresses, social security numbers, dates of birth, and medical treatment and health insurance information of 1,369,519 affected patients. Xsolis clarified that after the attack it had “implemented additional safeguards to [...] prevent similar incidents,” but its retrospective auditing reflects a security paradigm so widespread amongst AI providers that it resembles doctrine: for healthcare vendors, system failure is architectural red teaming.
Because of their non-deterministic nature and disparate ingestive data streams, deployed AI systems are mechanistically opaque, even to C-suite level executives. Vulnerabilities, in turn, are exposed primarily through failure: at a Tennessee hospital, Sentri7, a drug-monitoring software, failed to detect five instances of fentanyl diversion and regulators only uncovered the failure after doctors happened to observe an impaired nurse; UnitedHealthcare’s nH predict allegedly overrode physician decisions to deny medically necessary care at a 90% error rate, only disclosed after litigation – and only 0.2% of patients ever appeal.
These industry examples demonstrate that failure modes are exposed through crisis or litigation, not through transparent vendor reporting. Now, as AI systems begin to permeate entire healthcare networks, the risk threatens the architectural integrity of healthcare systems: 85% of healthcare organizations reported operational disruptions caused by third-party vendor failure last year. And while vendors can swallow cyber breaches as a mechanism to reveal system vulnerabilities, that paradigm is intolerable for providers: 61% of healthcare leaders believe a patient fatality will result from a cyberattack at a U.S. healthcare facility within the next five years.
References
[1] https://www.xsolisdataincident.com/
[2] https://cybernews.com/news/xsolis-humana-healthcare-data-breach-1-4-million-patients/
[3] https://www.securityweek.com/xsolis-data-breach-affects-1-4-million-individuals/
[5] https://www.forbes.com/sites/daraabasiita/2026/06/09/the-algorithm-that-counted-on-no-one-appealing/
[6] https://cybernews.com/news/xsolis-humana-healthcare-data-breach-1-4-million-patients/
[7] https://omegasystemscorp.com/insights/white-papers/2026-healthcare-it-landscape-report/