Security experts have found a 13-year-old flaw in Apache ActiveMQ Classic allowing hackers to remotely run code.

The vulnerability was discovered using the Claude AI, which found a way to exploit the system by analyzing how different parts work together.

Designated as CVE-2026-34197, this high-risk issue, with a severity score of 8.8, affects Apache ActiveMQ/Broker versions before 5.19.4 and versions 6.0.0 to 6.2.3.

This explains why it remained hidden for so long.

Apache ActiveMQ is a Java-based open-source tool that manages asynchronous communication through message queues.

Although ActiveMQ offers a newer, faster ‘Artemis’ version, the older ‘Classic’ version impacted by CVE-2026-34197 is widely used in enterprise systems, web infrastructure, government networks, and Java-based company setups.

Horizon3 researcher Naveen Sunkavally stated that he found the issue “with just a few simple instructions” to Claude. “Claude did 80% of the work, and a human refined it,” he added.

Sunkavally explained that Claude identified the problem by analyzing several components (Jolokia, JMX, network connectors, and VM transports).

“Each component works correctly on its own, but together they create a vulnerability. Claude excelled at connecting these elements seamlessly without preconceived notions.”

The researcher reported the vulnerability to Apache on March 22, and the developers released a fix on March 30 in ActiveMQ Classic versions 6.2.3 and 5.19.4.

A report from Horizon3 explains that the vulnerability is due to ActiveMQ’s Jolokia management API, which exposes a broker function (addNetworkConnector) that can be misused to load external configurations.

By sending a crafted request, an attacker can force the broker to load a remote Spring XML file and run system commands during startup.

While the issue requires Jolokia authentication, versions 6.0.0 to 6.1.1 are vulnerable to a separate bug, CVE-2024-32114, which exposes the API without any access control.

Horizon3 researchers stressed the danger of this new vulnerability, pointing out past ActiveMQ CVEs that have been exploited in actual attacks.

“We urge organizations using ActiveMQ to address this as a top priority, as ActiveMQ has frequently been targeted by attackers, and methods for exploiting it are well-known,” Horizon3 stated.

“CVE-2016-3088, an authenticated RCE in the web console, and CVE-2023-46604, an unauthenticated RCE in the broker port, are both on CISA’s KEV list.”

Although there are no reported instances of CVE-2026-34197 being actively exploited, the researchers indicate that signs of exploitation can be found in the ActiveMQ broker logs. They recommend searching for unusual broker connections using the internal transport protocol VM and the brokerConfig=xbean:http:// query parameter.

The command execution happens during multiple connection attempts. If a warning message about a configuration issue is displayed, the researchers say that the payload has already been executed.