Cisco BroadWorks and Openwave flaws exploited by autonomous AI agents
CISA adds two actively exploited vulnerabilities to its KEV catalog, including one used in an AI-driven campaign targeting telecom infrastructure.
The U.S. Cybersecurity and Infrastructure Security Agency (CISA) added two security flaws affecting Cisco BroadWorks and Openwave Messaging to its Known Exploited Vulnerabilities catalog on Tuesday. The agency cited evidence of active exploitation, including a sophisticated campaign driven by an autonomous AI agent that targeted a European telecom provider.
What happened
The first vulnerability, CVE-2026-20320, carries a CVSS score of 7.5. It is an improper restriction of XML external entity reference flaw in the Open Client Interface XML Parser of Cisco BroadWorks. This defect allows an unauthenticated, remote attacker to read sensitive configuration information with the privileges of the BroadWorks user. Cisco released a fix for this issue in August 2026.
The second flaw, CVE-2026-111793, has a CVSS score of 4.3. It involves incorrect resource transfer between spheres in the Openwave Messaging Webtop Media Transcoder. An authenticated user can exploit this to process and convert unintended document types if they bypass the Webmail experience. Japan Vulnerability Notes reported that the developer confirmed observed attacks exploiting this vulnerability, which was fixed in October 2026.
These additions follow recent disclosures from Cisco regarding active exploitation of vulnerabilities in Catalyst SD-WAN and Secure Firewall software. Federal Civilian Executive Branch agencies are now recommended to apply all relevant patches by November 24, 2026.
How it works
CVE-2026-20320 was leveraged in a campaign dubbed Operation Snowball by Malbear Labs. An unknown Chinese-speaking threat actor used an autonomous framework built on a jailbroken version of the DeepSeek Flash model. The AI agent targeted the BroadWorks eXtended Services Platform, specifically Release 23.0.1.1075 running on port 2208.
The agent exploited the unauthenticated XML external entity vulnerability to recover the system identity private key, three SSL credential sets, and three additional private keys. Possessing the system identity key allowed the attacker to impersonate the platform in inter-component communication. The agent then enumerated hundreds of assets, extracted dozens of credential sets, and exfiltrated thousands of rows of subscriber billing and device records over two weeks.
This autonomous system ran continuously, issuing parallel tasks to worker agents and adapting to defensive controls without human input. It used reconnaissance tools like Sliver and Chisel, along with web shells such as Behinder, to maintain persistent access. The operator only checked in asynchronously to review progress after the agent had completed its objectives.
Key details
- CVE-2026-20320 is an unauthenticated XXE vulnerability in Cisco BroadWorks OCI XML Parser, fixed in August 2026.
- CVE-2026-111793 allows authenticated users to process unintended document types in Openwave Messaging, fixed in October 2026.
- Operation Snowball utilized a jailbroken DeepSeek Flash model to autonomously target a European telecom provider.
- The AI agent exploited the BroadWorks flaw to steal system identity keys and SSL credentials.
- CISA recommends Federal Civilian Executive Branch agencies apply patches by November 24, 2026.
- The attack campaign also targeted vulnerabilities in products from JBoss, Liferay, ForgeRock, Linux, SAP, and others.
Why it matters
For engineering teams, this incident highlights the shifting landscape of automated threats. Attackers are no longer limited by human speed or attention spans. Autonomous agents can identify attack surfaces, choose exploitation paths, and adapt to defenses in real time. This means that known vulnerabilities with available patches become critical liabilities much faster than before. If a patch exists, it must be applied immediately, as AI-driven scanners will find and exploit unpatched systems continuously.
The specific nature of the BroadWorks exploit demonstrates the danger of exposed internal interfaces. The Open Client Interface XML Parser allowed unauthenticated access to sensitive configuration data. For developers building APIs or microservices, this underscores the importance of strict input validation and authentication checks on all endpoints, even those intended for internal use. Exposing such interfaces to the public internet without robust protection creates easy entry points for automated tools.
Furthermore, the theft of system identity keys illustrates the cascading impact of a single breach. Once the agent obtained the private key, it could impersonate legitimate components within the network. This lateral movement capability allows attackers to bypass traditional perimeter defenses. Teams must assume that any single compromised component can lead to full system impersonation, necessitating zero-trust architectures and rigorous key management practices.
What you can do
- Patch Cisco BroadWorks and Openwave Messaging systems immediately if you have not already applied the August and October 2026 updates.
- Audit your network for any other unpatched vulnerabilities listed in the CISA KEV catalog, particularly those with high CVSS scores.
- Review exposure of internal APIs and XML parsers, ensuring they are not accessible from untrusted networks without strong authentication.
- Implement strict key management protocols to limit the impact of stolen credentials or private keys.
- Monitor for unusual outbound traffic or reconnaissance activities that may indicate an autonomous agent is probing your infrastructure.
- Set up automated alerts for certificate and domain expiry to reduce the attack surface related to expired or misconfigured TLS assets.



