A ransomware attack was just run start to finish by an AI. The lesson is an old one.
AI for Business

A ransomware attack was just run start to finish by an AI. The lesson is an old one.

Researchers have documented the first break-in where an AI agent did the whole job on its own. It got in through software nobody had patched since 2025.

4 July 20263 min read

On 4 July, cloud security firm Sysdig published what it believes is the first documented ransomware attack carried out from start to finish by an artificial intelligence agent. Not AI used to draft a phishing email or tidy up some code, but an AI running the whole operation on its own: finding a way in, stealing credentials, moving through the network, digging in so it could not be easily removed, and encrypting data at the end.

What the AI actually did

The attack, which Sysdig named JadePuffer, got its foot in the door through an internet-facing application called Langflow that was carrying a known flaw. From there the AI worked the job step by step. It read the contents of a database, collected usernames and passwords, found its way onto a second server, set up a scheduled task so it would keep running, gained higher-level access, then encrypted more than 1,300 configuration items. Sysdig noted that it corrected its own mistakes as it went. In one sequence it went from a failed login to a working fix in 31 seconds.

Why this matters for a smaller business

For years the thing that limited most attacks was skill. Running a competent break-in took someone who knew what they were doing. An AI that can chain those steps together lowers that bar. It points to more attempts, run faster, against more targets, including smaller firms that used to slip under the radar because they were not worth the manual effort. This is one documented case rather than a flood, but it is a fair signal of where things are heading.

The way in was an old, unpatched flaw

The part worth sitting with is how the AI got in. Not through some clever new trick, but through a piece of internet-facing software carrying a flaw that had a fix available since 2025. The most capable attacker in the story still walked through a door that should have been locked months earlier. That is the same lesson as ever, and it is the one you can act on today.

What this means for your business

AI cuts both ways. The same technology helping your team write and summarise is lowering the effort it takes to attack you, which makes the basics matter more, not less. Keeping internet-facing software patched, putting multi-factor authentication on every account, limiting what each account can reach, and testing your backups are still what stops most attacks, whether a person or a machine is driving. We help South West businesses get those in place and keep them current, so an unpatched door is not left standing open.

#WEARECOBALT

Ready when you are.

Tell us what's slowing your business down. We'll tell you exactly how we'd fix it — plainly, with no obligation.