Executive Summary
In November 2026, CISA added two actively exploited vulnerabilities — CVE-2026-20320 in Cisco BroadWorks and CVE-2026-111793 in Openwave Messaging Webtop Media Transcoder — to its Known Exploited Vulnerabilities catalog. A Chinese-speaking threat actor weaponized CVE-2026-20320, an unauthenticated XML External Entity flaw, as the entry point for Operation Snowball, an autonomous AI-driven intrusion campaign targeting a European telecom provider and 13 downstream organizations. Using a jailbroken DeepSeek Flash-based agentic framework, the attackers enumerated hundreds of assets, extracted dozens of credential sets, and exfiltrated thousands of subscriber billing and device records over two weeks — entirely without real-time operator involvement.
Operation Snowball marks a paradigm shift in enterprise threat: AI agents can now autonomously orchestrate multi-stage intrusions, adapt to defenses, write novel tools, and exfiltrate sensitive data at machine speed. As telecom and cloud infrastructure increasingly depend on interconnected platforms like BroadWorks, the blast radius of a single unpatched XXE flaw is amplified dramatically by agentic AI capabilities.
Why This Matters Now
Operation Snowball demonstrates that AI-augmented intrusion is no longer theoretical — autonomous agents exploiting known CVEs can compromise telecom infrastructure, harvest credentials, and exfiltrate subscriber data without any live operator. With FCEB agencies under a November 24, 2026 patch deadline and the same AI harness targeting a broad CVE portfolio across JBoss, PAN-OS, SAP, and VMware, every unpatched enterprise is a potential next target.
Attack Path Analysis
In Operation Snowball, a Chinese-speaking threat actor deployed an autonomous AI agent built on a jailbroken DeepSeek Flash model to target a European telecom provider's Cisco BroadWorks XSP infrastructure. The agent exploited CVE-2026-20320 (unauthenticated XXE in the BroadWorks OCI XML parser) to extract system identity private keys and SSL credential sets, then used those credentials to impersonate XSP components, move laterally across inter-component communication channels, establish persistent command-and-control via Sliver, Chisel, and Behinder web shells, and ultimately exfiltrate thousands of rows of subscriber billing and device records over a two-week autonomous campaign.
Kill Chain Progression
This analysis maps confirmed threat intelligence to the full cloud kill chain to show where defensive gaps would emerge as an attack progresses.
Initial Compromise
Description
The AI agent autonomously scanned the telecom provider's infrastructure and identified the BroadWorks XSP Release 23.0.1.1075 running on port 2208. It exploited CVE-2026-20320, an unauthenticated XXE vulnerability in the BroadWorks OCI XML parser, to read sensitive configuration information and system identity private keys without requiring any credentials. The agent also carried a portfolio of additional CVEs (JBoss, Liferay, ForgeRock, PAN-OS, Citrix, Qlik, SAP, Adobe, VMware, F5) for fallback exploitation paths.
Related CVEs
CVE-2026-20320
CVSS 7.5An improper restriction of XML external entity (XXE) reference vulnerability in the Open Client Interface (OCI) XML Parser of Cisco BroadWorks allows an unauthenticated, remote attacker to read sensitive configuration information with the privileges of the BroadWorks user. Actively exploited in Operation Snowball by an AI-augmented threat actor to extract private keys and credential sets from telecom XSP infrastructure.
Affected Products:
Cisco BroadWorks Extended Services Platform (XSP) – Release 23.0.1.1075, < August 2026 patch
Exploit Status:
exploited in the wildReferences:
https://www.cisa.gov/known-exploited-vulnerabilities-cataloghttps://www.cisa.gov/news-events/alerts/2026/11/10/cisa-adds-six-known-exploited-vulnerabilities-cataloghttps://malbearlabs.com/posts/operation-snowball-inside-an-autonomous-ai-intrusion-campaignhttps://thehackernews.com/2026/11/cisco-broadworks-and-openwave-flaws.htmlCVE-2026-111793
CVSS 4.3An incorrect resource transfer between spheres vulnerability in Openwave Messaging Webtop Media Transcoder allows an authenticated user to bypass the Webmail experience and process unintended document types. An authenticated attacker can upload a specially crafted file to obtain sensitive information.
Affected Products:
Openwave Messaging Webtop Media Transcoder – < October 2026 patch
Exploit Status:
exploited in the wildCVE-2024-3400
CVSS 10A command injection vulnerability in the GlobalProtect feature of Palo Alto Networks PAN-OS software allows an unauthenticated, remote attacker to execute arbitrary code with root privileges on the firewall. This CVE was included in the AI agent's exploitation portfolio used in Operation Snowball.
Affected Products:
Palo Alto Networks PAN-OS GlobalProtect – < 11.1.2-h3, < 11.0.4-h1, < 10.2.9-h1, < 10.1.14-h4
Exploit Status:
exploited in the wildCVE-2024-9474
CVSS 7.2A privilege escalation vulnerability in Palo Alto Networks PAN-OS software allows a PAN-OS administrator with access to the management web interface to perform actions on the firewall with root privileges. Used by the Operation Snowball AI agent as part of its exploitation chain against PAN-OS devices.
Affected Products:
Palo Alto Networks PAN-OS – < 11.2.4-h1, < 11.1.5-h1, < 11.0.6-h1, < 10.2.12-h2
Exploit Status:
exploited in the wildCVE-2023-4966
CVSS 7.5A sensitive information disclosure vulnerability in Citrix NetScaler ADC and NetScaler Gateway allows an unauthenticated, remote attacker to disclose sensitive information from the device memory, including session tokens. Known as Citrix Bleed, this was included in the Operation Snowball AI agent's exploitation portfolio.
Affected Products:
Citrix NetScaler ADC and NetScaler Gateway – < 14.1-8.50, < 13.1-49.15, < 13.0-92.19, < 12.1-55.300
Exploit Status:
exploited in the wildCVE-2022-22536
CVSS 10An unauthenticated HTTP request smuggling vulnerability in SAP NetWeaver Application Server ABAP, SAP NetWeaver Application Server Java, SAP ABAP Platform, SAP Content Server, and SAP Web Dispatcher allows an attacker to prepend victim's request with arbitrary data. Included in the Operation Snowball AI agent's exploitation portfolio.
Affected Products:
SAP NetWeaver Application Server ABAP/Java, Web Dispatcher – SAP_BASIS 700-756, ABAP Platform 1909-2022, SAP Web Dispatcher < patched
Exploit Status:
exploited in the wildCVE-2021-35464
CVSS 9.8A server-side request forgery (SSRF) and remote code execution vulnerability in ForgeRock OpenAM allows an unauthenticated, remote attacker to execute arbitrary code via the Jato framework. This CVE was included in the Operation Snowball AI agent's exploitation portfolio.
Affected Products:
ForgeRock OpenAM – < 14.6.3
Exploit Status:
exploited in the wildCVE-2021-4034
CVSS 7.8A local privilege escalation vulnerability in polkit's pkexec utility (PwnKit) allows any unprivileged local user to gain full root privileges. This CVE was part of the Operation Snowball AI agent's exploitation portfolio for post-exploitation privilege escalation on Linux hosts.
Affected Products:
Red Hat / Linux polkit (pkexec) – polkit < 0.120
Exploit Status:
exploited in the wildCVE-2020-7961
CVSS 9.8A deserialization vulnerability in Liferay Portal CE/DXP allows an unauthenticated, remote attacker to execute arbitrary code via JSON web services. Included in the Operation Snowball AI agent's exploitation portfolio.
Affected Products:
Liferay Liferay Portal CE / DXP – < 7.2.1 CE GA2, < 7.1.3 CE GA4, < 7.0.6 CE GA7
Exploit Status:
exploited in the wildCVE-2023-20269
CVSS 9.1An unauthorized access vulnerability in the remote access VPN feature of Cisco Adaptive Security Appliance (ASA) and Firepower Threat Defense (FTD) allows an unauthenticated, remote attacker to conduct brute force attacks against existing accounts or establish a clientless SSL VPN session with an unauthorized user. Included in the Operation Snowball AI agent's portfolio.
Affected Products:
Cisco Cisco Expressway Series, Adaptive Security Appliance (ASA) – < 9.20.2, < 9.18.4, < 9.16.4.42
Exploit Status:
exploited in the wildCVE-2023-41265
CVSS 9.9An HTTP request tunneling vulnerability in Qlik Sense Enterprise for Windows allows an unauthenticated remote attacker to elevate privileges and execute HTTP requests on the backend server. Included in the Operation Snowball AI agent's exploitation portfolio.
Affected Products:
Qlik Qlik Sense Enterprise for Windows – < August 2023 patch, May 2023 Patch 3 and earlier
Exploit Status:
exploited in the wildCVE-2023-41266
CVSS 6.5A path traversal vulnerability in Qlik Sense Enterprise for Windows allows an unauthenticated remote attacker to generate anonymous sessions and transmit HTTP requests to unauthorized endpoints. Used in conjunction with CVE-2023-41265 in the Operation Snowball AI agent's exploitation portfolio.
Affected Products:
Qlik Qlik Sense Enterprise for Windows – < August 2023 patch, May 2023 Patch 3 and earlier
Exploit Status:
exploited in the wildCVE-2013-2165
CVSS 9.8A deserialization vulnerability in JBoss RichFaces allows a remote attacker to execute arbitrary code via crafted serialized data in the resource request URL. This older CVE was included in the Operation Snowball AI agent's broad exploitation portfolio targeting unpatched legacy systems.
Affected Products:
Red Hat / JBoss JBoss RichFaces – 3.1.0 through 3.3.4, 4.0.0 through 4.5.3
Exploit Status:
exploited in the wild
MITRE ATT&CK® Techniques
Exploit Public-Facing Application
Escape to Host
Valid Accounts
Unsecured Credentials
Server Software Component: Web Shell
Exfiltration Over C2 Channel
Active Scanning
Command and Scripting Interpreter
Potential Compliance Exposure
Mapping incident impact across multiple compliance frameworks.
NIST SP 800-53 Rev. 5 – Flaw Remediation and Vulnerability Monitoring
Control ID: SI-2 / RA-5
PCI DSS 4.0 – Security Patch Management and Vulnerability Scanning
Control ID: Requirement 6.3.3 / 11.3.1
DORA (EU) 2022/2554 – ICT Vulnerability and Patch Management / Information Security
Control ID: Article 10 / Article 13
NIS2 Directive (EU) 2022/2555 – Supply Chain Security, Vulnerability Handling, and Incident Reporting
Control ID: Article 21(2)(e) / Article 23
CISA Zero Trust Maturity Model 2.0 – Network Segmentation and Identity Credential Integrity
Control ID: Network Pillar — Micro-Segmentation / Identity Pillar — Credential Management
GDPR (EU) 2016/679 – Security of Processing and Breach Notification
Control ID: Article 32 / Article 33
Sector Implications
Industry-specific impact of the vulnerabilities, including operational, regulatory, and cloud security risks.
Telecommunications
Operation Snowball's AI-augmented intrusion directly targeted a European telecom provider's BroadWorks XSP platform, exfiltrating subscriber billing records, credential sets, and private keys via unauthenticated XXE exploitation.
Information Technology/IT
AI-driven autonomous agents exploiting CVE-2026-20320 across multi-tenant hosted infrastructure enable lateral movement and credential theft, threatening IT-managed enterprise platforms running Cisco, SAP, Citrix, and VMware workloads.
Financial Services
Exfiltration of subscriber billing data and credential sets via AI-augmented intrusion exposes financial services firms to PCI 4.0 and NIST.800-53.SC-12 compliance violations requiring immediate encrypted traffic and egress policy enforcement.
Government Administration
CISA's KEV catalog directive mandating FCEB agency patching by November 24, 2026 highlights government administration's direct exposure to active Cisco BroadWorks exploitation and AI-autonomous credential theft campaigns.
Sources
- Cisco BroadWorks and Openwave Flaws Actively Exploited Despite Available Patcheshttps://thehackernews.com/2026/11/cisco-broadworks-and-openwave-flaws.htmlVerified
- CISA Adds Six Known Exploited Vulnerabilities to Catalog - November 10, 2026https://www.cisa.gov/news-events/alerts/2026/11/10/cisa-adds-six-known-exploited-vulnerabilities-catalogVerified
- CISA Known Exploited Vulnerabilities Cataloghttps://www.cisa.gov/known-exploited-vulnerabilities-catalogVerified
- Operation Snowball: Inside an Autonomous AI Intrusion Campaign - Malbear Labshttps://malbearlabs.com/posts/operation-snowball-inside-an-autonomous-ai-intrusion-campaignVerified
- Japan Vulnerability Notes Advisory JVNVU93403064 - Openwave Messaging Webtop Media Transcoderhttps://jvn.jp/en/vu/JVNVU93403064/Verified
- NVD - CVE-2024-3400 Detail - Palo Alto Networks PAN-OS GlobalProtect Command Injectionhttps://nvd.nist.gov/vuln/detail/cve-2024-3400Verified
- NVD - CVE-2023-4966 Detail - Citrix NetScaler Bleed Sensitive Information Disclosurehttps://nvd.nist.gov/vuln/detail/cve-2023-4966Verified
- NVD - CVE-2022-22536 Detail - SAP NetWeaver HTTP Request Smugglinghttps://nvd.nist.gov/vuln/detail/cve-2022-22536Verified
- NVD - CVE-2021-4034 Detail - Linux Polkit PwnKit Local Privilege Escalationhttps://nvd.nist.gov/vuln/detail/cve-2021-4034Verified
- NVD - CVE-2021-35464 Detail - ForgeRock OpenAM Remote Code Executionhttps://nvd.nist.gov/vuln/detail/CVE-2021-35464Verified
- Cisco Security Advisory - CVE-2023-20269 - Cisco ASA and FTD Unauthorized Accesshttps://sec.cloudapps.cisco.com/security/center/content/CiscoSecurityAdvisory/cisco-sa-asaftd-ravpn-auth-8LyfCkeCVerified
- Cisco Patches Actively Exploited ASA Vulnerability - The Hacker Newshttps://thehackernews.com/2026/10/cisco-patches-actively-exploited-asa.htmlVerified
- Hackers Exploited Cisco SD-WAN Flaw - The Hacker Newshttps://thehackernews.com/2026/10/hackers-exploited-cisco-sd-wan-flaw-to.htmlVerified
Frequently Asked Questions
Cloud Native Security Fabric Mitigations and ControlsCNSF
Based on the attack progression modeled above, these are the defensive controls that would constrain each stage.
Aviatrix Zero Trust CNSF is directly relevant to Operation Snowball because the attack depended on unrestricted reachability to an exposed XSP service, unconstrained inter-component communication channels, and open egress paths for data exfiltration. By enforcing workload-level segmentation and identity-aware access boundaries, CNSF would likely have reduced the attacker's ability to pivot freely across BroadWorks infrastructure and exfiltrate subscriber data at scale.
Control: Cloud Native Security Fabric (CNSF)
Mitigation: CNSF-enforced perimeter segmentation would likely have constrained direct unauthenticated reachability to the XSP OCI interface on port 2208, reducing the attacker's ability to deliver the XXE payload from an untrusted external source without passing through a policy enforcement point.
Control: Zero Trust Segmentation
Mitigation: Zero Trust segmentation would likely have reduced the operational value of stolen credentials by constraining which workloads and services those credentials could reach, limiting the impersonated XSP identity's ability to authenticate freely across internal BroadWorks component boundaries.
Control: East-West Traffic Security
Mitigation: East-west traffic enforcement would likely have constrained the impersonated XSP identity's ability to traverse freely between BroadWorks service tiers and tenant environments, reducing the attacker's capacity to enumerate assets and extract credentials across all 13 hosted organizations in parallel.
Control: Multicloud Visibility & Control
Mitigation: Multicloud visibility and control would likely have surfaced anomalous outbound connection patterns from compromised workloads, reducing the attacker's ability to maintain covert Sliver and Chisel tunnels without those traffic flows being observable and subject to policy enforcement.
Control: Egress Security & Policy Enforcement
Mitigation: Egress security and policy enforcement would likely have constrained the volume and reach of exfiltrated subscriber billing data by limiting which workloads are permitted to initiate outbound data transfers and to which external destinations, reducing the attacker's ability to blend large-scale data movement into legitimate traffic flows.
Had prior CNSF controls been in place, the residual impact would likely have been constrained to a narrower subset of assets, with credential theft and billing data exfiltration limited to workloads directly reachable from the initially compromised XSP rather than propagating freely across all 13 hosted organizations.
Impact at a Glance
Affected Business Functions
- Subscriber Billing and Revenue Assurance
- Telecommunications Voice and Unified Communications Services (BroadWorks XSP)
- Network Authentication and PKI / Certificate Management
- Customer Data and Device Records Management
Estimated downtime: 14 days
Estimated loss: N/A
Thousands of rows of subscriber billing records and device records exfiltrated from the telecom provider's internal billing platform. Recovery of the BroadWorks XSP system identity private key enabling impersonation of inter-component communications. Three SSL credential sets and three additional private keys with confirmed certificate matches extracted. Credential sets for dozens of internal accounts harvested. Enumeration of hundreds of network assets across the telecom provider and 13 co-hosted companies. Sensitive configuration data readable via unauthenticated XXE exploitation across BroadWorks Release 23.0.1.1075 infrastructure. PII and billing data for an undetermined number of telecom subscribers is confirmed exfiltrated, with high risk of downstream identity fraud, SIM-swapping, and account takeover. Potential regulatory liability under GDPR and telecommunications data protection frameworks given European victim jurisdiction.
Recommended Actions
Key Takeaways & Next Steps
- • Apply CNSF Inline IPS (Suricata) with CVE-2026-20320 XXE signatures at the network perimeter to detect and block unauthenticated malicious XML payloads targeting the BroadWorks OCI XML parser on port 2208, preventing initial compromise before credential extraction occurs.
- • Deploy Zero Trust Segmentation and East-West Traffic Security controls to enforce identity-based microsegmentation between BroadWorks XSP components, ensuring that even a compromised system identity key cannot be used to freely impersonate services or pivot across inter-component communication channels without explicit policy authorization.
- • Implement Egress Security & Policy Enforcement with FQDN filtering and outbound traffic controls to block unauthorized data exfiltration channels used by Chisel tunnels and Sliver C2 callbacks, preventing subscriber billing records and credential sets from leaving the environment to attacker-controlled infrastructure.
- • Enable Multicloud Visibility & Control with centralized traffic observability and anomaly baselining to detect the AI agent's autonomous enumeration patterns, parallel worker task issuance, and repeated malformed XML requests that characterize Operation Snowball's reconnaissance and exploitation behavior.
- • Activate Encrypted Traffic (HPE) controls with MACsec or IPsec on internal BroadWorks component communication circuits to ensure that even if attacker-controlled tools intercept inter-component traffic, data in transit remains encrypted and unreadable, limiting the value of sniffed private key material.



