13 Years to Find It. 9 Days to Exploit It. Minutes to Patch It.

On March 30, 2026, Apache shipped patches for CVE-2026-34197, a remote code execution bug in ActiveMQ Classic that had been sitting in the codebase for thirteen years. Horizon3.ai had reported it to the project on March 22. Apache had a fix out in eight days.

On April 7, Horizon3.ai's Naveen Sunkavally published the public disclosure.

On April 16, CISA added it to the Known Exploited Vulnerabilities catalog. Federal agencies have until April 30 to patch.

Seventeen days from fix to active exploitation. Nine days from public disclosure. ShadowServer is currently tracking more than 7,500 ActiveMQ servers exposed online, and most of the downstream supply chain still hasn't shipped a remediated image.

That's the story. It's also the only story that matters in 2026.

The find

Sunkavally didn't spend a week reverse-engineering ActiveMQ. He pointed Claude at the source code with, in his words, "a couple of basic prompts." The AI traced a chain through four components that were developed independently over the last decade: Jolokia, JMX, network connectors, and the in-process VM transport.

The path: Jolokia's default access policy permits exec operations on every ActiveMQ MBean. One of those operations, BrokerService.addNetworkConnector(String), accepts a discovery URI. Feed it a vm:// URI with a brokerConfig=xbean:http parameter, and the broker loads a remote Spring XML application context via ResourceXmlApplicationContext. Spring instantiates every singleton bean before ActiveMQ validates the config. Bean factory methods reach Runtime.exec(). Arbitrary code execution on the broker's JVM.

Sunkavally's own estimate: work that would have taken him a week manually, Claude did in ten minutes. He called the split 80% Claude, 20% human gift-wrapping.

The exploit

The attack surface is worse than the CVSS 8.8 suggests. On ActiveMQ 6.0.0 through 6.1.1, an unrelated bug (CVE-2024-32114) already exposed the Jolokia endpoint without authentication. On those versions, CVE-2026-34197 is effectively an unauthenticated RCE. And where authentication is required, admin:admin is still the default most deployments never got around to changing.

ActiveMQ Classic is not a niche product. It sits inside enterprise backends, government systems, and a long tail of vendor appliances that bundled a Java message broker a decade ago and never looked at it again. None of those are patching themselves today.

FortiGuard telemetry shows exploitation attempts peaking on April 14, two days before CISA added the CVE to KEV. Defenders waiting for the CISA trigger were already behind.

CISA gave federal agencies 14 days. Attackers gave them zero. The math is not subtle.

The patch gap

Apache shipped fixes in ActiveMQ Classic 5.19.4 and 6.2.3 on March 30. That was the easy part. Now look at the delivery pipeline most teams actually depend on:

1. Upstream cuts a patched release. (Days.)

2. A base-image vendor or distro maintainer (bitnami, the Apache image, Debian, Ubuntu, RHEL, SUSE) picks up the fix, rebases it against the specific ActiveMQ version they ship, tests it, and pushes it through their release process. (Weeks. Sometimes months. Sometimes never, for old stable branches.)

3. Vendor appliances with embedded brokers rebuild against the updated package. (Days after upstream.)

4. You rebuild your application images on the new base. (Whenever CI gets to it.)

5. The update rolls through staging into production. (Whenever change-management lets it.)

As of this writing, most packaged distributions of ActiveMQ have not picked up the fix. If you are running a base image that bundles it, or a vendor appliance with an embedded broker, you are waiting.

The asymmetry

This is not an ActiveMQ problem.

Researchers using AI to find bugs is not the interesting part of this story. Attackers using AI to find and weaponize bugs is. The same workflow that took Sunkavally ten minutes against ActiveMQ can be pointed at every Spring dependency, every npm tree, every forgotten Java library buried three layers deep in a vendor image. The flaw hid for thirteen years because it required whole-codebase reasoning across four independently-developed components. That is exactly the kind of assumption-free analysis LLMs are now good at. The marginal cost of a new vulnerability discovery is collapsing toward zero.

Discovery used to be the bottleneck. It isn't anymore. The bottleneck is the patch-distribution pipeline. The weeks-to-months machinery that assumed vulnerabilities would arrive at a human-reviewable cadence.

That assumption is gone, and nothing in the traditional stack is built for what's replacing it. If discovery is AI-accelerated and fix-delivery still moves at human-maintainer speed, defenders lose every cycle. That's not a prediction. That's the structure of the 17-day window CISA just documented.

What to do today

If you run ActiveMQ Classic, in any form:

1. Inventory every instance. Containers, VMs, bare metal, vendor appliances with embedded brokers. Include the dusty ones. ShadowServer says 7,500+ are exposed right now. Don't assume you aren't one.

2. Upgrade to 5.19.4 or 6.2.3 where you can. Upstream fixes are the cleanest path.

3. Restrict /api/jolokia/ to trusted networks. Don't rely on authentication alone. Default credentials are too common to trust.

4. Hunt for indicators of compromise in your broker logs:

   • Network connector activity referencing vm:// URIs with brokerConfig=xbean:http

   • POST requests to /api/jolokia/ containing addNetworkConnector in the request body

   • Outbound HTTP requests from the ActiveMQ broker process to unexpected hosts

   • Unexpected child processes spawned by the ActiveMQ Java process

5. If you cannot upgrade. Your distro hasn't shipped a backport, your vendor won't, or the version you need is out of support. Start the free trial. We'll have a patched image for you before your next standup.

What Root does

This is the gap we built Root to close.

When a CVE like this drops, we don't wait for a base-image rebuild or a vendor appliance refresh. We take the exact ActiveMQ version you are running, whatever release, whatever distro, whatever vendor appliance, and produce a patched container image in minutes. Our agentic remediation pipeline handles the source-to-binary mapping, applies the fix, rebuilds, validates, and ships a versioned image that slots cleanly into your deployment.

By the time a base-image vendor files the backport ticket, we have shipped you an image of the ActiveMQ version you are running, patched and ready.

Time from CVE disclosure to a patched image in your hand: minutes to hours. Not weeks.

This is the same model that put us ahead of the TeamPCP campaign in Q1, when Trivy, LiteLLM, axios, and Checkmarx KICS were compromised in a five-day cascade. Root had remediated versions of every affected package ready in minutes.

Try it. Root's free trial is open. Full product access. Container images and application libraries. Automatic provisioning. No strings. Running through May 31.

If you are running ActiveMQ Classic anywhere in your stack, a patched image is a signup away.

Start here: app.root.io