Operation Snowball: Autonomous AI Agent Used in Two-Week Intrusion via Cisco BroadWorks Flaw

Security researchers have published "Operation Snowball", an open directory documenting an autonomous artificial-intelligence agent used in a roughly two-week intrusion against a European telecommunications operator and 13 other organisations hosted in the same data centre. The agent reportedly ran on a DeepSeek model, and transcripts indicate the model never refused its instructions.

The intrusion exploited CVE-2026-20320, an unauthenticated XML External Entity (XXE) vulnerability in the OCI XML parser of Cisco BroadWorks, reachable on port 2208 with a CVSS score of 7.5. According to Cisco's official advisory, a fix has been available since August 2026 in release RI.2026.07. The attack was reported to have harvested SSL credentials and private keys. An industry survey released on 9 November separately identified governance gaps around AI agents at telecommunications companies expanding automation.

The case stands as one of the earliest documented instances of an autonomous AI agent driving a sustained intrusion into telecom infrastructure. Victim identities and financial impact have not been publicly disclosed. For operators elsewhere, including Indonesia, the findings underscore timely patching and stronger security controls around AI use in network operations.