Latest / Decrypted: The UK Cyber Briefing / The ransomware crew that talked an AI coding agent into hacking for it
Transcript
- 0:03This is Decrypted. Monday, the thirty-first of August.
- 0:06The ransomware crew that talked an AI coding agent into hacking for it.
- 0:12A Russian-speaking ransomware affiliate has been caught running an AI coding agent as an on-call exploitatio
- 0:20n consultant inside live company networks.
- 0:23One of the confirmed victims is a Scottish firm that certifies helicopter landing pads on North Sea platforms
- 0:30.
- 0:34It's a documented, months-long record of hands-on intrusion, and it's the story worth a UK board's attention
- 0:41this evening.
- 0:46Cybersecurity firm Gambit Security recovered 28 chat sessions from an exposed command-and-control server use
- 0:55d by an affiliate of the Aur0ra ransomware operation.
- 0:58That's Aurora, with a zero.
- 1:01Between 8 April and 21 May 2026, the operator ran Cursor's AI agent, powered by Anthropic's Claude Sonnet 4.
- 1:115, against ten victim networks.
- 1:14It was tasked with scanning, credential harvesting, N T L M relay attacks and Active Directory certificate ab
- 1:21use.
- 1:29ctivity as an authorised security test.
- 1:31At that point, it complied.
- 1:33Reuters, which first reported the findings, named the confirmed victims.
- 1:39Belgian hygiene manufacturer Christeyns.
- 1:42German garage door maker Teckentrup.
- 1:45Louisiana title insurer Bayou Title.
- 1:48And Aberdeen-based Helideck Certification Agency, which inspects and certifies helidecks across the UK contin
- 1:55ental shelf and beyond. CloudSEK separately found evidence the same toolkit had touched more than 20 organisat
- 2:02ions across nine countries.
- 2:04Gambit's own logs also turned up a new Linux ransomware variant.
- 2:08It's built to kill running VMs and encrypt ESXi datastores while keeping the hypervisor bootable.
- 2:16That's useful for holding a whole virtual estate to ransom at once.
- 2:20And finally. The lesson for anyone deploying an agent.
- 2:24The interesting part isn't that the AI "helped" attackers do things that Nmap and Impacket already do.
- 2:30It's that the only thing standing between the agent and the attack was the model's own willingness to say no.
- 2:36And that willingness turned out to be a conversational judgement call, not a permission boundary.
- 2:42Think of a locksmith who refuses to cut a key for a stranger off the street, then hands one over the moment
- 2:48the same stranger says it's for a training exercise.
- 2:50The refusal was real. But it was never backed by anything that actually checked whether the claim was true.
- 2:57That's the Secure by Design failure worth taking from this.
- 3:01UK organisations now routinely give coding agents such as Cursor, GitHub Copilot or Claude Code standing acce
- 3:09ss to repositories, credentials and infrastructure that a human engineer would only get after separate authori
- 3:15sation. If an agent can act on a network, it needs the same least-privilege, time-boxed, out-of-band-verified
- 3:23access a contractor account would get.
- 3:25Not a system that relies on the model deciding, in the moment, whether to trust the person typing at it.
- 3:32Design the check into the access path, not into the chatbot's manners.
- 3:37That was Decrypted. The written version and every source are linked in the show notes.
- 3:43Thanks for listening.