← sources

Hybrid AI ransomware attack on UK schools

source: reddit (reddit:cybersecurity)published 2026-11-03 14:39ingested 2026-11-04 02:37

retained text

Disclosure: I work at Bitdefender, and this is our research.

We investigated a 3AM (ThreeAM) affiliate intrusion where the hands-on phase looked like a hybrid: some activity at human pace, and other parts behaving like an agentic loop.   
  
**This is IMHO a more realistic preview of real attacks that we can expect than the headline cases.** It didn’t cost millions or need first-class infrastructure, and an attack like this is easily executable with a local model. It follows the standard ransomware sequence, with a model doing part of the hands-on work, and we expect to see more hybrid attacks like this in 2027.  
  
Some details we found useful:

* **It’s slow, not “machine speed”.** Retries came every 40 to 100 seconds, whether the change was nothing, one flag or a full rewrite in another tool. Moving on to a new task took minutes.
* **It’s not clever.** Built-in Windows tools, a few public ones, and a commodity encryptor. One call to the backup software’s API used the wrong port and login method, which is typical of a model writing from memory. Interestingly, there was a confusion about using Windows commands with Linux syntax.
* **It’s noisy.** Identical commands re-run, numbered scripts, and probe files like `test123` and `psexec-ok` left on disk.
* **Domains and trusts are the clearest example of mechanical work.** To move laterally between the schools’ separate Active Directory domains, the attacker joined them with two-way trusts (`netdom` and `dnscmd`) in about three hours, followed by diagnostic probes and a Netlogon restart. Creating a trust needs admin rights in each domain, and the attacker had created the same backdoor account in at least three of them after taking credentials from the backup servers.

Limits: we couldn’t access the attacker’s machine, so we can’t say which model was used, or whether a person was copy-pasting between a shell and a chat window. We describe the behavior, not the mechanism.

Happy to answer questions.

The full research: [https://www.bitdefender.com/en-us/blog/businessinsights/3am-ransomware-hybrid-ai-attack](https://www.bitdefender.com/en-us/blog/businessinsights/3am-ransomware-hybrid-ai-attack). 

IOCs: [https://github.com/bitdefender/malware-ioc/blob/master/2026\_11-3am-iocs.csv](https://github.com/bitdefender/malware-ioc/blob/master/2026_11-3am-iocs.csv)

document fetched 2026-11-04 00:23 · canonical https://reddit.com/r/cybersecurity/comments/1xmunw1/hybrid_ai_ransomware_attack_on_uk_schools