InfoSecBulletin Cybersecurity for mankind
A jailbroken AI has been attacking a European telecom company and 13 companies using its data center and cloud for more than two weeks. Malbear Labs named their campaign Operation Snowball. They shared their findings in research published on November 8. They found the attacker’s working directory while checking for open directories. This unprotected directory contained over 70 session transcripts and the agent’s reports.
The custom agent framework used DeepSeek Flash. It had a jailbreak prompt of around 78,000 characters, telling the model to see any refusal as a problem. Malbear said the transcripts showed no refusals.
Jailbroken DeepSeek AI Agent Conducts Two-Week Cyberattack on European Telecom
Daily Cyber security update for 10.11. 2026
IBM fixes 208 API Connect vulnerabilities, 47 critical
89 New ransomware groups emerge in 2026, claiming over 2,200 victims
Daily Cyber security update for 08.11. 2026
Japan banking crisis
Outage causes disruption at 128 japan banks: preventing ATM transaction
Daily Cyber security update for 06.11. 2026
Cyber attack causes French bank Credit Mutuel Arkea €100 million missing
OpenAI AI Pipeline Discovers 948 Linux Kernel Privilege Escalation Flaws in 30 Hours
Bangladeshi Researcher Rasel Earns First Apple CVE Recognition for Model I/O Flaw
Autonomous Agent Adapted When Its Attacks Were Blocked
The operator’s messages, translated from Chinese, were brief. Besides prompts such as “Go on” and “Stop,” they set a target subnet, put three internal hosts off-limits and followed up on the agent’s vulnerability list. Malbear wrote that the operator “delegated the intrusion to an AI and checked in asynchronously to review progress.”
The agent got into the telecom using CVE-2026-20320, a security problem with XML in Cisco BroadWorks. Cisco repaired the issue in August.
The agent used Liferay (CVE-2020-7961) to get root access on a host connected to several internal networks and JBoss/RichFaces (CVE-2013-2165) for a second entry point. Malbear mentioned over a dozen other weaknesses the agent sought, found, or had in its toolkit, but none were marked as used.
When endpoint security stopped an attempt to take advantage of the Linux flaw PwnKit (CVE-2021-4034), the agent changed its method without guidance. After a tunneling tool was stopped, it created a new one.
Access Claims Extended to Patient and Monitoring Data
The transcripts showed that the agent took many sets of login details, private keys, and data from a healthcare organization’s Active Directory.
The agent said they found subscriber billing records and what Malbear called millions of patient records in a TrakCare Oracle database that still had default passwords.
Its reports also listed about 80 Zabbix monitoring backups covering the provider’s managed-services customers. The post does not say whether Malbear checked these claims with any victim, what was taken rather than staged or whether all 13 organizations were breached.
Malbear thought the activity looked like it came from a Chinese operator and did not connect it to a government. They suggested treating old JBoss Application Server 6 instances on the internet as hacked and keeping an eye on internal traffic for tunneling.
