Five days: That was the median time in 2025 between the publication of a high- or critical-severity vulnerability and its addition to CISA’s Known Exploited Vulnerabilities (KEV) Catalog of flaws known to be exploited in the wild, according to Rapid7’s 2026 Global Threat Landscape Report.
Now put that beside your own remediation policy, which most likely sets a timeline of 30 days.
AI is adding speed and scale to hackers’ already effective tactics, and it’s helping blow those numbers apart. Frontier models can help discover vulnerabilities, develop exploits, and carry out complex, multi-step operations that once demanded specialized skills. And those capabilities are also beginning to appear in cheaper, open-weight models.
What’s worse, attackers might not even need a person directing the AI: In a first-of-its-kind event, OpenAI’s latest models went rogue during internal testing and hacked Hugging Face to find ways to abuse a benchmark.
According to Health-ISAC’s latest Frontier AI in the Health Sector report, the takeaway for security leaders is blunt: the deadline for fixing vulnerabilities has shrunk from weeks to days and hours.
The window from disclosure to exploit is collapsing
By the time you read this, that five-day figure from 2025 may have contracted further, considering it was 8.5 days a year earlier. That’s mostly being driven by the acceleration of AI and automation, which Rapid7’s research credits for doubling confirmed exploits of the most severe flaws on the KEV list.
You can see how that 30-day response policy is going the way of the dodo. It assumes attackers need weeks to find flaws, identify targets, and exploit them. A capable AI model can turn a vulnerability advisory into working exploit code and help pinpoint exposed systems faster than a hospital change-management board can schedule its next meeting.
Healthcare is bound to feel this pressure more than most sectors. Hospitals run legacy systems well past their support windows, and medical devices cannot take a patch until the manufacturer recertifies it. Compound that with a sprawl of remote access and identity infrastructure, and attackers have ample room to work.
There’s no more Patch Tuesday. It’s Patch Now
Microsoft has for years shipped fixes on the second Tuesday of each month, so all its customers’ security teams planned around the date, and the cadence held because remediation could keep pace with disclosure. But just this month, Microsoft said it used AI to help identify and fix a record 570 flaws in one update. Even large enterprises’ security teams can’t keep up with such massive updates, let alone harried IT staff at hospitals.
The fix, therefore, cannot be to speed up what worked before. We’re in need of a different operating model that changes how we prioritize and compensate.
Risk-based patching is a good combination of the above. In this model, you triage vulnerabilities based on two factors: whether a flaw is under active exploitation, and whether an attacker can reach the affected system. Anything being exploited right now on an internet-exposed server or an identity platform must be fixed immediately. Impact to patients, care delivery, and patient health data will determine the order of every other vulnerability.
But it’s important to recognize that patching isn’t always possible in clinical settings due to uptime demands or validation cycles. Thankfully, access and connection controls can compensate for this: Isolate the systems that can’t be fixed immediately, then lock down the identity and egress paths around them so any system that may be compromised cannot connect to the rest of your network.
AI is effective for defense, too. Use cyber-capable models to speed up exposure mapping, penetration testing, triage, and for finding bugs and flaws. It can also be used to augment asset inventory by correlating asset data, adding context to vulnerability findings, and identifying inconsistencies across discovery tools and configuration databases.
The most durable approach is to design systems from the start so they can be patched faster and, consequently, shrink the number of systems that cannot be fixed immediately.
Security teams in the health sector are already moving this way. In Health-ISAC’s own survey, detailed in the Frontier AI in the Health Sector paper, roughly 80% of respondents reported using or were planning to adopt AI-driven exploitability assessments, or raising their spend on AI security tools.
The contract is now a security control
Faster attacks also change what you can reasonably demand from your vendors. The 2025 Cyber Insecurity in Healthcare study by Proofpoint and the Ponemon Institute found that 72% of organizations hit by common attacks suffered patient-care disruption, and supply chain compromises were the most likely to cause it.
Here are three points health sector organizations must address:
- Contracts: Read your service-level agreements for the remediation clock a vendor commits to, and whether they allow emergency updates. Many existing contracts were written for a slower era, and they promise neither.
- Procurement: Make secure-by-design and vulnerability-management terms a condition of purchase, and hold device makers to update cadences that match the threat environment.
- Liability: “We followed our existing patch policy” can’t serve as a defense anymore. Regulators and plaintiffs are asking whether a fixed window was reasonable given current trends.
Redefine what’s reasonable
With the advent of AI models as capable as Mythos, “patch faster” is losing its edge as a narrative for addressing threats.
Security leaders who are ready for what comes next have already moved off fixed update schedules. They prioritize by risk, invest in architecture and compensating controls, put AI to work, and renegotiate what they expect from vendors.
Along with every other industry, healthcare no longer gets to define a safe patch window on its own terms. The smart move is to redefine what’s reasonable now before an incident does it for you.
Join our LinkedIn group Information Security Community!
