3AM Affiliate Explores AI-Assisted Operations in European School Attack

A recent ransomware attack targeting a UK school estate, attributed to a 3AM affiliate, may offer an early glimpse into how cybercriminals could leverage artificial intelligence as a force multiplier rather than a complete replacement for human operators. Bitdefender's research highlights what they describe as a "hybrid AI ransomware attack," where certain actions appeared to be executed by agentic elements rather than solely by human hands.

The attack itself followed a largely familiar ransomware playbook. The affiliate successfully infiltrated a school, subsequently utilizing standard Windows tools to establish domain trusts, gain access to other schools within the estate, exfiltrate data, and deploy the 3AM ransomware. Key tactics included stealing backup credentials, creating rogue domain administrator accounts, and establishing trust relationships between previously separate Active Directory domains.

What distinguished this incident was the observed command generation and troubleshooting process. Bitdefender's analysis suggests that an AI system may have been responsible for handling repetitive command generation and dynamic revision during the breach, while a human operator maintained oversight for high-level decision-making. This is not a fully automated AI attack, nor a purely human-driven one, but a blend of both.

Evidence supporting the AI involvement includes unusual command patterns, such as calls to the Veeam backup software's web interface with parameters that deviate from official documentation. Bitdefender noted mistakes typical of an AI generating commands from its training data rather than consulting a manual. The presence of self-labeled output files, numbered script revisions, and retry behaviors resembling a loop further support this hypothesis.

Researchers also observed that the time intervals between PowerShell command attempts were shorter than a human would typically execute manually, yet longer than what a fully autonomous LLM might produce. This suggests a human selecting the next step while an AI handles the iterative execution and retries, a pattern Bitdefender characterizes as "hybrid."

While Bitdefender could not directly access the attacker's machine to confirm the AI model's specifics, the observed behaviors strongly indicate a human operating in conjunction with an AI agent. This "hybrid model" is considered more technically and economically feasible for financially motivated cybercriminals than a fully autonomous attack.

The implications extend beyond this single incident. The attack targeted a critical sector, and the affiliate's ability to scale operations by automating repetitive tasks could be replicated against other similar institutions. The attacker's reconnaissance efforts, including enumerating trust relationships and searching for other schools within the network, highlight the potential for widespread impact.

Defenders can counter such hybrid attacks by adhering to standard ransomware prevention strategies. These include maintaining separate administrative credentials across domains, treating backups as a high-value target, restricting remote execution and administrative shares, blocking unnecessary interpreters and shells, and ensuring robust endpoint security solutions are deployed across all machines. The repetitive nature of agentic loop actions also provides potential indicators for detection.

This new report details a specific instance of a 3AM ransomware affiliate targeting UK schools, expanding on the previously observed AI-like command generation and dynamic revision. The attack involved exploiting Veeam backups, using Mimikatz, and creating fake admin accounts, resulting in 37 encrypted machines. Researchers noted the hybrid human-AI approach, with AI agents potentially automating repetitive tasks like retrying failed commands, a pattern observed in other recent attacks.