AI Made Finding Vulnerabilities Easy. Fixing Them Is Now the Job

By
Emanuel Salmona
November 12, 2026
•
6
min read
Share this post

JPMorgan Chase has long been one of the most thoughtful voices in cybersecurity, and this month its technology leaders shared something every security team should read.

Their Global Technology Leadership Team published 200 days into the frontier: What we know and what to do now, a letter to CEOs about what frontier AI means for cyber risk. I've read it several times now, and I'm grateful they chose to share their thinking so openly. The central claim is simple: AI has made finding and exploiting weaknesses much faster, and the work of fixing them has not sped up to match.

When we started Nagomi, that gap was the problem we wanted to work on. It was real then. It is wider now, and one of the most respected security organizations in the world is asking every chief executive to take it seriously. Here's what I took from the post, and what I think security leaders should do with it.

Discovery got faster. Remediation didn't.

For most of the last twenty years, finding vulnerabilities took real effort. Scanners improved, bug bounties grew, and researchers got better, but discovery still had a natural speed limit, and so did attackers.

That limit is going away. Models can now read code, reason about how systems connect, and find flaws at a pace no human team can match. The JPMorgan Chase team makes the point that attackers get the same advantage.

The fixing side hasn't changed much. Patching still means testing, change windows, owners who are hard to find, and systems no one wants to touch. So the backlog grows, and the time between a flaw becoming known and being used shrinks. Their recommendation is for companies to reduce the risk they've built up over years, rather than waiting for a patch to arrive for each problem.

Deciding fast is not the same as acting fast

The part of the post I keep coming back to is a distinction most organizations blur. One is decision speed: how quickly leadership can approve an emergency change. The other is execution speed: how quickly the organization can actually find every affected system, reach its owner, and put a fix or a control in place.

Many companies have worked hard on the first. Incident playbooks, executive escalation paths, pre-approved emergency changes. Those matter. But as the post points out, fast approval doesn't help if the asset data is incomplete or the fix can't be rolled out at scale.

In our conversations with security teams, this is where the time goes. Not in the meeting where someone says yes, but in the days before it, spent working out what's affected and who owns it, and in the weeks after, spent chasing the fix across teams.

Three questions should decide what gets fixed first

If every vulnerability is urgent, none of them are. The post's advice is to stop treating every finding the same and instead ask, for each asset:

  1. How reachable is it, especially from the internet?
  2. Can it still be protected? Does the vendor still ship security fixes, and are there controls in front of it?
  3. What happens to the business if it's compromised?

None of these questions is new. What's new is that the answers have to come in hours, and for most organizations they still come from spreadsheets, tribal knowledge, and a lot of messages asking who owns what.

The second question is the one I'd point security leaders to first, because it's where a lot of existing investment goes unused. Most companies already own good firewalls, endpoint protection, identity controls and web application firewalls. The question isn't whether they have tools. It's whether those tools are deployed and configured to protect the specific assets that matter, and whether anyone can prove it.

Compensating controls only count if you can show them

The post offers clear guidance on end-of-support technology: treat it as a liability with a deadline. Every exception should have a named owner, compensating controls, and a date for getting off the old system.

That last requirement is harder than it sounds. "We have a compensating control" is one of the most common sentences in security, and one of the least checked. Controls drift. Agents fall off machines. A rule that was right a year ago gets loosened for an outage and never tightened again. We see this in our own research: in Nagomi's 2026 Enterprise Exposure Snapshot, The Illusion of Maturity, three in four organizations had gaps in core controls such as multi-factor authentication, endpoint detection and policy enforcement, even when their vulnerability scores looked strong.

This connects to the thread that runs through every recommendation in the post: leaders should ask for evidence, not assurance. I think that's the most important line in it for security teams, because the board is going to start asking. When they do, "we're on it" won't be enough. You'll need to show which exposures exist, which are covered by controls that are actually working, and which are not.

Why we built Nagomi around this problem

We started Nagomi because we kept seeing the same thing: capable security teams with good tools, buried in findings, unable to answer simple questions quickly. Which of these actually matter? Are we already protected? Who needs to do what?

Adding another source of findings doesn't fix that. What helps is connecting what you already have, your assets, your exposures and the controls you've paid for, so you can see which risks are already covered, which aren't, and what to fix first. Then you can route the work to the people who own it through the workflows they already use.

JPMorgan Chase has invested for years to build the people and practices this takes, and the post reflects that experience. Most companies are earlier on that path. They need the same answers with the team they have today, and that's where we want to help.

A test to run this quarter

The post closes with a challenge for CEOs: picture a serious vulnerability surfacing tomorrow, and name what would keep your company from responding the same day. For a lot of organizations, I suspect the honest answer is a control they assumed was working.

Here's a test that takes an afternoon. List the three compensating controls your team relies on most: the firewall rule in front of a legacy system, the endpoint agent on servers you can't patch, the WAF policy protecting an old application. Then check each one today. Is it deployed everywhere it should be? Is it configured the way you think it is? When was it last verified, and by whom?

If all three hold up, you have real evidence to bring to your board. If even one doesn't, you've found the gap before an attacker did, and you know where to start.

Our thanks to the JPMorgan Chase team for sharing what they've learned so generously. It's the kind of openness that makes the whole industry stronger, and it's worth acting on.

Read the full post from JPMorgan Chase: 200 days into the frontier: What we know and what to do now.

Table of contents