Skip to the content
AI and Cybersecurity Learning and Practice Center Test automation Open the Learning Center

IncidentAI-enabled attacks

Bitdefender: a 3AM ransomware affiliate encrypted 37 machines at a UK school estate, with commands that may have come from an AI loop

HighBitdefender3AM

BleepingComputer reports Bitdefender's investigation of an attack by a 3AM ransomware affiliate on a UK school estate between September 10 and 19. The attacker pulled credentials from three Veeam backup servers with a modified public PowerShell tool, copied a Chrome credentials database, ran Mimikatz and LSASS dumping tools, created two backdoor domain admin accounts named library and svc_bkpmon, stole financial data, set up two-way trusts between three separate Active Directory domains, tried to remove Sophos and Bitdefender protection, deleted system-state backups and encrypted 37 of nearly 50 machines reached from an unmanaged host. The initial access could not be established. Bitdefender believes the attacker may have used AI-generated commands and an automated agentic loop, since the same commands came back slightly changed every 40 to 100 seconds, but it could not examine the attacker's machine, so it cannot document anything about a model, and a person pasting between the implant and a model would look the same.

Replay the attack step by step in the Learning Center

How it happened

  1. The attacker harvested credentials from backup servers, then moved across domains it had linked with new trusts.
  2. Repeated, slightly altered commands at 40 to 100 second intervals are why Bitdefender suspects an automated AI loop.

Impact

37 machines were encrypted and financial data was stolen from the school estate.

The fix

The article reports no advice from Bitdefender. Practical steps that follow from the steps described: keep backup servers and their stored credentials isolated, alert on new domain admin accounts and new domain trusts, and protect endpoint security tools from being switched off.

Lessons

  • Backup servers that store credentials are a shortcut to the whole network; isolate them.
  • A new domain admin or a new trust between domains should raise an alert at once.
  • Suspected AI use is a hypothesis from timing; it changes the pace of an attack more than its method.

Sources