AI is expanding the attack surface for hospitals faster than governance can respond

Hospitals, health systems, physician practices, payers, medtech companies and pharmaceutical companies are racing to adopt AI, competing for the efficiency and innovation it promises. Cybersecurity and governance haven’t kept up. Healthcare organizations are now challenged to recognize and mitigate the risks created by that gap.

AI is creating new ways to breach hospitals 

AI is enhancing threat actors’ capabilities, and it is introducing new vulnerabilities into the tech stacks of the healthcare organizations that use it. A new report from Black Book Research identifies several ways in which AI expands the attack surface for hospitals. Hospitals could be breached via user prompts and data uploaded to AI tools and systems, the AI models and applications they use, retrieval-augmented generation (RAG) and knowledge stores, AI agents and the tools they have access to and third-party AI vendors.

Malicious actors can target any part of an organization’s AI stack, but there is internal risk as well. The proliferation of shadow AI threatens the ability to monitor and control the movement of sensitive data.

Chirag Shah, global information security officer and data privacy officer at revenue management software company Model N, offered the examples of employees using unapproved AI tools to summarize medical records or draft clinical documentation.

“Sensitive information can move outside approved systems. Business logic can be exposed and regulated data may be processed in ways that are hard to monitor,” he said.

Third-party risk is getting more complicated

Vendors eager to capitalize on AI innovation are integrating the technology into existing tools and launching new offerings. Hospitals and other healthcare organizations shoulder the risk that comes with using tools that have access to their systems and sensitive data.

“Security is going to be relevant not just with respect to the IT infrastructure and networks that are controlled by the healthcare provider or healthcare company, but the security of the infrastructure and network security of the AI provider or the AI model provider is going to be just as relevant,” Ryan Sulkin, team lead, data privacy and cybersecurity practice group at Benesch Law, said.

More than half of healthcare providers (56%) surveyed by healthcare infrastructure company Cotiviti and MedCity News report being moderately concerned that their AI vendors are introducing cybersecurity and compliance vulnerabilities. An additional 24% of providers are extremely concerned.

The pursuit of innovation is winning over governance

The cybersecurity risks of AI are not unknown, although finding where that risk lies in an organization and figuring out what to do about it is another matter. But good cybersecurity and overall governance take time, and the pressure to capture innovation and competitive edge is an immediate demand.

“The fastest AI adoption often happens in the places where governance is least mature,” Shah said. And when that happens, the cybersecurity risk is high.

Even when healthcare organizations recognize the need for more mature governance, they have to grapple with resource constraints. Nancy Phillips, CISO at Ensemble Health Partners, a revenue cycle management company, has observed the perception among some organizations that AI, which promises its users greater efficiency through automation, is the solution to resource constraints.

“It’s going to take a little more time to see some true partners in that space that will then start to help organizations alleviate some of that pressure from a resource standpoint,” she said. “But it’s just not there yet.”

In the meantime, healthcare organizations are adding more AI tools to their tech stacks, many of which are designed for cybersecurity. But the Black Book Research report emphasizes the distinction between AI cybersecurity tools and securing AI systems. The former might help accelerate vulnerability management or threat detection, but that doesn’t mean they also address the risk of AI tools leaking data or acting in unexpected ways.

Risk lives in the gaps between AI adoption and mature cybersecurity controls 

Cybersecurity for AI requires visibility, continuous monitoring and accountability. Without mature cybersecurity controls, healthcare organizations risk data leakage, operational disruption and negative patient outcomes. A hardened perimeter is no longer enough, according to Phillips.

“We can’t even have trust on the inside,” she said. “Everything has to be explicitly allowed or denied in order to make sure that these solutions are doing only what they’re supposed to be doing and nothing more.”

The entire industry is currently in the early days of a steep learning curve, attempting to get cybersecurity and governance up to speed while regulations are just as nascent.

“The absence of modern law for these issues means we’re going to fall back on older law,” Sulkin said. “Whenever you’re applying older law to new technology, there’s a lot of room for litigation.”

Mayo Clinic is the subject of a whistleblower lawsuit that claims the health system concealed errors and put patient safety and data at risk in its push to adopt AI.

Litigation like this and new regulatory oversight will take time to shape the way responsible and safe AI use is defined. In the meantime, healthcare organizations have to find a way to narrow the gap between innovation and risk.

This doesn’t necessarily mean slamming the brakes on AI adoption in healthcare, according to Sulkin. “What it requires is a partnership between the innovators at a company and the risk managers at a company, which is not always the easiest thing to do,” he said. “But in my experience, it can be done effectively.”

Carrie Pallardy, a Chicago-based freelance writer and editor, began her career covering healthcare more than a decade ago. Her work has taken into many different industries, but covering healthcare delivery remains a constant focus. She can be reached at [email protected] or on LinkedIn.