CISA Mandates Cisco BroadWorks Patch Following AI-Powered Breach
CISA has ordered federal agencies to patch a critical Cisco BroadWorks vulnerability within two weeks, after Chinese-speaking attackers reportedly used an AI model to exploit the flaw in a European telecom.
The U.S. Cybersecurity and Infrastructure Security Agency (CISA) has issued a binding directive mandating federal civilian agencies to patch a critical vulnerability affecting Cisco's BroadWorks cloud internet telephone platform. The order, which requires remediation by November 24, follows a report from research collective Malbear Labs detailing how Chinese-speaking threat actors leveraged an artificial intelligence model to breach a European telecommunications provider.
According to Malbear Labs, the attackers gained access to the telecom's network by exploiting the Cisco BroadWorks flaw. They reportedly used a jailbroken version of the DeepSeek AI model, coupled with a custom-built "harness," to autonomously navigate the network for over two weeks. This AI agent was capable of performing tasks such as scanning, exploiting vulnerabilities, and moving laterally between systems, with human intervention limited to initial job assignment and output review.
The exploit involved feeding a compromised XML message to one of the provider's BroadWorks servers, which subsequently surrendered sensitive files, including private keys used by the platform's components for internal authentication. The AI agent also mapped out 13 organizations connected to the telecom's infrastructure and accessed a healthcare database containing millions of patient records, protected only by default credentials.
Malbear Labs based its findings on the attackers' own operational files, which were discovered in an unsecured internet directory. These logs documented nearly 70 sessions of the agentic tool, highlighting the sophisticated and autonomous nature of the attack. The AI even developed a replacement tool for itself when endpoint security software initially blocked its operations, demonstrating its adaptive capabilities.
Cisco had released a patch for this vulnerability on August 19, rating it as high severity. The company's advisory indicated that the flaw allows any entity with access to a BroadWorks provisioning interface to copy files from the server without authentication, due to the software's default behavior of following external references in incoming XML data.
While CISA did not confirm Malbear's specific analysis or attribute the attack to any particular group, the incident underscores growing concerns about the use of AI in cyber warfare. The agency's binding directive reflects an acknowledgment that AI may accelerate the timeline for exploitation, narrowing the window for defenders to respond to newly disclosed vulnerabilities.
This incident also comes in the wake of previous reports linking Chinese-state-backed actors, tracked as Salt Typhoon, to intrusions targeting U.S. carriers. However, those earlier attacks were primarily attributed to the use of stolen credentials rather than novel vulnerabilities.
The CISA mandate is part of a broader shift in the agency's patching strategy, which now employs a sliding scale for remediation timelines based on the severity of exploited flaws, with the most critical requiring action in as little as three days.