← All tech posts

Satya Nadella Just Said the Quiet Part Out Loud: Critical Infrastructure Is 'Unpatchable'

Satya Nadella doesn't usually say the thing everyone in security already knows but companies don't put in press releases. This week, at The New York Times' DealBook Summit, he did.

"Remember, there is a lot of stuff in our critical infrastructure that's unpatchable," Nadella said. "I mean, think about all the firmware and the chillers and the HVAC systems everywhere else. So all of that, by the way, they're on the internet."

That sentence landed differently than it would have two years ago, because of what had just happened the week before: six separate US outages between November 12 and 17, hitting Microsoft's business email, US air traffic control, several major banks, United Airlines, Apple, and the NSA's own classified network. Investigators concluded most of those were self-inflicted — five of the six operators broke their own systems, and the FAA still couldn't say what caused its failure. Only one incident, through a tool called Optimizely that big companies run on their websites, was a confirmed attack. But the timing made Nadella's "unpatchable" comment land like a confession rather than a warning.

The chillers are on the internet, and nobody's updating them

Nadella's examples weren't abstract. He pointed specifically at building cooling systems — chillers, the kind that keep data centers from cooking themselves — and HVAC systems. These are devices that got "smartified" over the last decade: given network interfaces, remote monitoring, IoT connectivity, all in the name of efficiency. Nobody designed them assuming they'd need a security patch cadence, because for most of their existence they didn't need one. Now they're reachable from the internet, running firmware nobody at the facility knows how to update, built by a vendor who may not exist anymore.

"Somebody was telling me about sort of even the transportation systems and the amount of stuff that's discoverable but with unpatchable firmware," Nadella added. Discoverable is the key word. In the pre-AI era, finding which of a million internet-connected chillers had a known firmware CVE was tedious, manual, specialist work. It's not anymore.

AI didn't just speed up attacks. It broke the patching timeline that used to work

The more interesting part of Nadella's comments wasn't about unpatchable legacy hardware — it was about what happens to the stuff that can be patched. "We now have a very powerful tool that finds vulnerabilities," he said. That's true for defenders and attackers both, and it's collapsed the buffer that used to exist between "a flaw is found" and "a flaw is exploited."

Under the old model, a company had time: test a patch, understand the blast radius, stage a rollout, deploy carefully. Nadella's point is that window is gone. "The patch now needs to go instantly from being available to being patched in production, because it's being exploited in the wild, right? So that's going to create instability."

Read that sentence as a security engineer and it should worry you more than a new vulnerability class would. He's describing a forced trade-off between speed and stability, at the scale of critical infrastructure, with AI compressing both sides of the equation — attackers moving faster, and defenders forced to skip the testing rigor that used to prevent bad patches from becoming their own outage. Several analysts covering the remarks compared the moment to a slower-motion, harder-to-predict Y2K: not a single deadline everyone prepared for, but a rolling, permanent compression of the time between discovery and exploitation, with no fixed date to plan around.

Why this matters beyond one CEO's soundbite

Nadella's comments arrived in a week already saturated with AI-and-infrastructure anxiety. Reports surfaced of an AI-generated cyberattack targeting a key EU institution in October, with some of the bloc's digital tools reportedly still offline weeks later. Separately, Australia announced legislation forcing AI firms to report breaches, after an OpenAI agent was linked to a Medicare data incident. The pattern across all of it is the same: AI is lowering the cost of both finding flaws and exploiting them, in systems that were never built to be found or exploited at this speed.

The uncomfortable implication for anyone running infrastructure — not just data centers, but hospitals, water systems, transit, manufacturing — is that "patch everything" was never actually achievable, and the gap between what's patchable and what isn't is now a live attack surface that's easier than ever to map. Nadella's prescription was modernization plus real-time detection: assess what's actually discoverable, assume some of it can never be fixed in place, and build the ability to detect and contain an attack in the moment rather than relying on prevention alone.

That's a much less comfortable message than "buy our security product," which may be exactly why it's the one that's gone viral this week — a tech CEO, in front of a room of executives, admitting that a meaningful share of the systems running modern life simply cannot be fixed, and that AI just made that fact dangerous in a way it wasn't twelve months ago.

Find what attackers can see, before they do.

AI pentesting agents test your web app, API and cloud, and every finding comes with evidence and a fix. First report in about 10 minutes.

Stay ahead of attackers.

New vulnerabilities, breach write-ups and practical fixes. A few emails a month.