"The only truly secure system is one that is powered off, cast in a block of concrete and sealed in a lead-lined room.”
                            Bill Gates

I am gathering notes for upcoming talks and consultations on Cybersecurity in Healthcare in the AI Era. This domain is perhaps just as white hot a topic as AI in healthcare in general. In this era of AI, these two areas of AI and cybersecurity are becoming increasingly more intertwined for multiple stakeholders. Cybersecurity was once a technical subject that boards delegated to others but recent developments in AI will end that strategy. Artificial intelligence became concomitantly the attacker's and the defender's new tools in health systems. AI and cybersecurity have climbed to the top of the agendas for boards of directors and C suites. 

These notes are divided into two halves. Part I is a plain-language primer on the ten threats every executive and director should know. Part II suggests a cybersecurity strategy for a hospital that is now AI-enabled.

Part I: A Cybersecurity Primer for Executives and Directors
Why hospitals are targets

Hospitals combine three elements attackers find attractive: data that cannot be changed (a diagnosis, unlike a credit card number, cannot be reissued), operations that cannot stop, and technology that is hard to update. Health care has been the costliest industry for data breaches for 13 consecutive years, at an average of $6.64 million per incident in IBM's 2026 report. One tracker counted over 400 ransomware attacks on healthcare organizations worldwide in the first half of 2026, more than two a day.
The harm is clinical as well as financial. When systems go down, ambulances divert, surgeries are postponed and staff return to paper with high inefficiencies and time lost. The 2024 attack on Change Healthcare exposed data on about 190 million people and froze claims payment across the country. Cybersecurity is a patient safety matter, and directors should treat it as any urgent risk: a risk that is never eliminated, but only managed.
 

 The top ten threats
The top ten threats below are briefly described and overlap as real attacks can chain several of these threats together: a phishing email yields a password, the password opens a vendor portal, and ransomware then follows. Each entry gives a definition, the consequence for any hospital or health system, and one question a director can put to AI leadership:
 

1. Ransomware. Criminals break in, lock (encrypt) the hospital's files and systems, and demand payment for the key. Most now steal data first and threaten to publish it, a tactic called double extortion. For a hospital this is the threat that stops care: the record system, lab, pharmacy and imaging can go dark for weeks if not months. Paying does not guarantee recovery, and the larger cost is usually the rebuilding and lost revenue that follow. Ask: how long could/should we deliver safe care on paper, and when did we last rehearse it?

2. Phishing. A deceptive message tricks a person into clicking a link, opening a file or giving up a password. It remains the most common way in, because the attacks are aimed at people, not machines. AI has unfortunately sharpened this attack: messages are now fluent and personal, and a voice can be cloned from a few seconds of audio to impersonate an executive or a physician. Variants include "vishing" (voice phishing) by phone and "smishing" by text message (SMS). Ask: can our help desk reset a password on the strength of a phone call alone?

3. Malware. The umbrella term for malicious software: viruses, "trojans" hidden inside legitimate-looking programs, "infostealers" that harvest saved passwords, and "wipers" that destroy data. Ransomware is one kind. It arrives through email, infected websites, USB drives or compromised software updates. Modern malware often hides for weeks while it maps the network, which is why detection matters as much as prevention. Ask: do we watch every computer and server for suspicious behavior, not just for known bad files?

4. Stolen credentials. Many intruders do not break in; they log in, with a password that was stolen, guessed or bought. One analysis found that 82% of intrusions in 2025 involved no malware at all. The most effective single defense is multi-factor authentication (MFA), which demands a second proof of identity such as a phone prompt or a physical key. The Change Healthcare attack began at a remote-access portal without it. Ask: is MFA on every remote entry point and every privileged account, without exception?

5. Third-party and supply chain risk. A hospital depends on hundreds of vendors: the record system, billing clearinghouses, cloud hosts, device makers. An attack on any one vendor becomes the hospital's problem: the Change Healthcare attack, on a single claims clearinghouse, disrupted payment for most providers in the country. Vetting a vendor at signing is not enough; the risk needs watching for the life of the relationship. Ask: which five vendors could halt our operations, and what is our plan if each goes dark tomorrow?

6. Medical device vulnerabilities. Infusion pumps, monitors, imaging machines and building systems are all computers on the network. Many run old software, cannot easily be updated, and stay in service for a decade or more. These can be an entry point or, in the worst case, be manipulated directly. Responsibility is often split between clinical/biomedical engineering and IT, and gaps often open between them. Ask: do we have a complete inventory of connected devices, and are they walled off from the rest of the network?

7. Insider threats. These are harm caused by people with legitimate access: the employee who snoops in a chart, the departing staffer who copies files and, and far more often, the well-meaning clinician who makes a mistake. A new variant is "shadow AI," in which staff paste patient information into unapproved AI tools. Most insider incidents are errors, so the remedy is design and training more than discipline. Ask: do people have access only to what their job requires, and would we notice unusual use?

8. Unpatched and legacy systems. Software flaws are discovered constantly, and vendors issue fixes called patches. Attackers now scan for a newly announced flaw within minutes, and AI is shortening the interval between disclosure and attack. A flaw exploited before any fix exists is a "zero-day." Hospitals struggle here because clinical systems cannot be taken offline casually and old systems no longer receive fixes. Ask: how many days do critical patches take us, and which unsupported systems are we still running?

9. Service disruption attacks. A distributed denial-of-service (DDoS) attack floods a website or network with traffic until it fails. Attackers use a large group of internet-connected devices call a botnet. It rarely touches patient data, but it cuts off portals, scheduling and telehealth, and it can serve as a diversion for a quieter intrusion. Some politically motivated groups have aimed such attacks at hospitals during international conflicts. Ask: are our public-facing services protected, and can we reach patients when these are down?

10. Data theft and exposure. Not every breach is dramatic. Records leave through a misconfigured cloud folder, a lost laptop, a misdirected email or a tracking tool on a website. A medical record holds everything needed for identity theft and insurance fraud, and each breach brings notification duties, regulatory scrutiny and litigation. Under HIPAA, a breach affecting 500 or more people must be reported to federal regulators and announced publicly within 60 days. Ask: do we know where all our patient data lives, including copies held by vendors and AI tools?
 

What good defense looks like
Just like terrorist attacks in the real world, no hospital can prevent every attack. Good cybersecurity programs share three cybersecurity tactics. First, they layer defenses so that no single failure is fatal, an approach called defense in depth. They assume a breach will occur, and so limit how far an intruder can move (segmentation) and how much any one account can do (least privilege). Second, they practice recovery, with tested backups kept beyond an attacker's reach and downtime drills run with clinicians. The standard reference is the NIST Cybersecurity Framework, organized around six verbs: govern, identify, protect, detect, respond, recover. Third, they deploy voluntary cybersecurity performance goals for hospitals published by the federal health department, a useful minimum checklist.

Resilience and recovery
Prevention measures with probably eventually fail, so two further measures matter as much as defense. Resilience is the capacity to keep delivering safe care while under attack. Recovery is the return to normal operations afterward. Both are measured in time. The recovery time objective is how long a system may be down before patients are harmed. The recovery point objective is how much recent data the hospital can afford to lose. Leadership should set both for each critical system, starting with the record system, pharmacy, laboratory and imaging.
Four practices make those numbers real. Backups are kept in several copies, at least one of them offline or unalterable ("immutable"), and restoration is tested, since an untested backup is only a hope. Downtime procedures let clinicians work on paper for weeks, not hours. An incident response plan names who decides, who speaks publicly, and when to call regulators, law enforcement and insurers. The hospital should know the order in which systems come back. Recovery from a major attack takes weeks for even months: the 2024 attack on Ascension kept electronic records offline at many of its hospitals for several weeks.
Directors and executives need not master the technology. Directors and executives do need to know, however, which of these cybersecurity measures the hospital has, which it lacks, and what closing the gap will cost.
Next time, we will discuss a strategic map as well as timeline for improving cybersecurity in healthcare for any healthcare organization.