Cisco BroadWorks Vulnerability Exploited in AI Attack Campaign
The Cisco BroadWorks vulnerability CVE-2026-20320 is being exploited in an attack campaign that used an autonomous artificial intelligence system against a European telecommunications provider and 13 companies hosted on its infrastructure. The U.S. Cybersecurity and Infrastructure Security Agency (CISA) added the flaw, along with a separate Openwave Messaging Webtop Media Transcoder vulnerability, to its Known Exploited Vulnerabilities (KEV) catalog after citing evidence of active exploitation.
CVE-2026-20320 has a CVSS severity score of 7.5. It is an improper restriction of XML external entity reference, or XXE, in the Open Client Interface XML Parser used by Cisco BroadWorks. XML is a format for organising data, while an external entity reference can make a parser retrieve information from another location. In this case, an unauthenticated remote attacker could read sensitive configuration information with the privileges of the Cisco BroadWorks user. Cisco fixed the issue in August 2026.
The other entry, CVE-2026-111793, has a CVSS score of 4.3 and affects Openwave Messaging Webtop Media Transcoder. It is an incorrect resource transfer between spheres, allowing an authenticated user to process and convert unintended document types by bypassing the Webmail experience. The issue was fixed in October 2026. Japan Vulnerability Notes said in an advisory that the developer had observed attacks exploiting the vulnerability. The portal also warned that an authenticated attacker could upload a specially crafted file and obtain sensitive information.
The BroadWorks exploitation was part of Operation Snowball, according to Malbear Labs. The activity was attributed to an unknown Chinese-speaking threat actor who used an autonomous framework built with a jailbroken version of the DeepSeek Flash model. The campaign targeted a European telecom provider and 13 companies hosted on its infrastructure. Over two weeks, the activity enumerated hundreds of assets, extracted dozens of credential sets, and exfiltrated thousands of rows of subscriber billing and device records from the provider’s internal billing platform.
The primary target was a BroadWorks XSP, or eXtended Services Platform, running Release 23.0.1.1075 on port 2208. Malbear Labs said the agent exploited the unauthenticated XXE in the BroadWorks OCI XML parser. From the XSP host, it recovered the system identity private key, three SSL credential sets, and three additional private keys with confirmed certificate matches. The system identity key is used by BroadWorks components to authenticate to one another, so possession of it allows an attacker to impersonate the XSP during communication between those components.
The framework also used reconnaissance tools, Sliver for post-exploitation activity, Chisel, and the Behinder web shell for persistent access. Malbear Labs described an intrusion in which the operator delegated work to an AI system and reviewed progress asynchronously. The primary agent ran continuously, assigned parallel tasks to worker agents, maintained an operational plan, adapted to defensive controls, created new tools when existing ones failed, extracted credentials, exfiltrated data, and compiled a finished target list without the operator being present during those actions.
The framework reportedly carried a portfolio of known vulnerabilities affecting JBoss RichFaces (CVE-2013-2165), Liferay CE/DXP (CVE-2020-7961), ForgeRock OpenAM (CVE-2021-35464), Linux (CVE-2021-4034), SAP NetWeaver (CVE-2022-22536), Cisco Expressway (CVE-2023-20269), Citrix NetScaler (CVE-2023-4966), Qlik Sense (CVE-2023-41265 and CVE-2023-41266), PAN-OS GlobalProtect (CVE-2024-3400 and CVE-2024-9474), Adobe Experience Manager (CVE-2025-54249), Broadcom VMware Cloud Director (CVE-2026-2003, CVE-2026-2004, CVE-2026-2005 and CVE-2026-2006), Claris FileMaker Server (CVE-2026-86934), and F5 BIG-IP (CVE-2026-41225). The list shows why applying available fixes matters even when a product is not the campaign’s main target: automated systems can select from many known weaknesses.
Cisco also disclosed late last month that CVE-2026-20127 in Catalyst SD-WAN, with a CVSS score of 10.0, and CVE-2026-76418 in Secure Firewall Adaptive Security Appliance Software and Secure Firewall Threat Defense Software, with a CVSS score of 8.6, were being exploited in the wild. Federal Civilian Executive Branch agencies were advised to apply the relevant patches by November 24, 2026. For organisations running affected products, the immediate priorities are confirming product versions, applying the vendor fixes, reviewing authentication material associated with BroadWorks systems, and checking for unusual access or data transfers.
For organisations that need broader help securing infrastructure around websites and business systems, AEU-I provides security-first IT, infrastructure and consulting. That service does not change the need to apply the relevant vendor patches and investigate possible compromise.