Cybersecurity researchers at Bitdefender detected, for the first time in Europe, a ransomware attack displaying behavioural indicators of the use of artificial intelligence (AI) agents during the infiltration and operation stages within the victim's network.

According to a company release sent to AGERPRES on Wednesday, the attack, attributed to an affiliate of the group behind the 3AM ransomware threat, targeted a group of schools in the UK, with an infrastructure made up of 47 systems spread across several separate domains.

Experts say this is the first documented case of this kind in Europe.

'The investigation shows that the attackers followed a sequence typical of ransomware operations: theft of access credentials from backup servers, creating unauthorised administrator accounts, connection of separate domains, exfiltrating data, attempting to disable security solutions and encrypting systems. The distinguishing feature is the way the tasks were divided. Some activities indicate direct intervention by a human operator, while others display a pattern consistent with the use of an AI agent that executes commands, interprets the results and adjusts its approach until the operation succeeds', the release informed.

Cybersecurity experts have also identified several indicators which, taken together, support the hypothesis that AI agents were used, including the presence of mistakes typical of commands generated from the AI model's learned knowledge.

'The pace of operations also highlights two distinct working patterns. Important decisions were separated by pauses of several minutes, consistent with human intervention. By contrast, attempts to fix failed operations were repeated at intervals of between 40 and 100 seconds, regardless of the scale of the modification. The clearest example was the connection of the schools' separate domains. This stage involved repetitive configurations, where the fix comes from reading an error and trying again, in a cycle consistent with the way an AI agent operates. The attackers stole credentials from backup servers and created two unauthorised administrator accounts, with names intended to blend in with a school's list of accounts. The attempt to remove the security solution from workstations was blocked by anti-tampering protection mechanisms. On 19 September, approximately 37 systems were encrypted. During the attack, system backups were also deleted', the quoted source states.

Bitdefender estimates that hybrid attacks, in which human operators coordinate the activities of AI agents, will become more frequent in 2027.

'At the same time, the use of AI agents leaves traces that can help with detection. Repeated commands, numbered scripts and test files left behind in the infrastructure provide security teams with more opportunities to identify malicious activity. The conclusions regarding the use of AI are based on the behaviour observed in the compromised network. The researchers did not have access to the attackers' systems and cannot determine which AI model was used or whether a human operator manually transferred commands between the AI tool and the targeted infrastructure. The research continues Bitdefender's efforts to document the use of artificial intelligence in cyberattacks, following its analysis of AI-generated code in the APT36 case, published in March 2026, and AI-assisted development in the SilkParasite case, from August 2026. The new case points to the expansion of AI use from creating attack tools to carrying out stages of the attack itself,' the specialists mentioned.

Bitdefender is a recognised leader in IT security, providing advanced prevention, detection and response solutions for cybersecurity incidents. Founded in 2001, the company has customers in more than 170 countries and offices on every continent. AGERPRES (RO - writing by: Daniel Badea; EN - writing by: Rodica State)