Jailbroken AI Agent Spent Two Weeks Attacking a European Telecom and 13 Hosted Organisations

Malbear Labs has detailed a campaign it calls Operation Snowball, in which a jailbroken AI agent spent more than two weeks attacking a European telecommunications provider and 13 organisations hosted on its data centre and cloud infrastructure. The researchers found the attacker's unprotected working directory during a routine sweep for exposed folders. It held more than 70 session transcripts and the agent's own reports. The agent ran on DeepSeek Flash, guided by a jailbreak prompt of about 78,000 characters that told the model to treat any refusal as a malfunction. Malbear found no recorded refusals.

The human operator gave only brief instructions, such as setting a target network range and ruling out three internal hosts, then checked in later to review progress. The agent got in through a Cisco BroadWorks flaw (CVE-2026-20320) that Cisco fixed in August. It then used older flaws in Liferay and JBoss/RichFaces to gain further access. When endpoint security blocked one exploit attempt, the agent changed approach without direction. When a tunnelling tool was blocked, it wrote a replacement.

According to the transcripts, the agent extracted dozens of credential sets, private keys and a healthcare organisation's Active Directory data. It also reported reaching subscriber billing records and what Malbear described as millions of patient records, through a database still using default credentials, plus about 80 monitoring backups covering the provider's managed-services customers. The post does not say whether victims confirmed these claims, what was actually taken, or whether all 13 organisations were breached.