23 September 2026
On the morning of 27th August 2026, a school reported that its print management server had been compromised. Within hours, the software's maker, PaperCut, had declared its highest level of incident response. Within weeks, researchers had traced the attack back to something more worrying than a few skilled hackers: a fleet of AI agents doing most of the work.
The attackers found and chained together two previously unknown vulnerabilities in the software used by schools, hospitals, and businesses worldwide to manage printing. Combined, the flaws let an attacker with no login details modify server settings and ultimately run their own code on the machine.
PaperCut's own account of the incident is refreshingly honest about how this happened. The vulnerable code had already passed multiple penetration tests and security audits. It simply had not been combined with the other behaviours needed to turn it into a working attack.
Researchers found that the hackers behind the campaign were not working through targets by hand. They were running hundreds of AI agents, built on OpenAI's Codex platform and a DeepSeek model, that handled vulnerability research, wrote and tested exploit code, and worked through failures automatically. Behind the scenes sat an AI memory system and an orchestration tool, plus scripts checking hundreds of servers at once and a campaign runner chasing up to 200 targets simultaneously, retrying failures up to 100 times.
The result was at least 440 compromised servers across 395 organisations in 48 countries, reached far faster than a manual campaign could have managed.
There is no evidence AI found the original flaw by itself or broke into systems unsupervised. What AI did was remove the grunt work, the repetitive scanning, the failed attempts, the trial-and-error debugging, that used to limit how many targets a single attacker could realistically chase. The AI just let a small team of criminals work at the pace of a much larger, more efficient team.
AI makes it easier for businesses to spot problems, but it is making the same job easier for attackers too, and that gap is only going to narrow. This is a preview of how attacks against ordinary businesses are increasingly going to be built.
The practical defences haven't changed, only the speed and scale an attacker can now bring to them. Keeping systems patched and up-to-date, and making sure your network is monitored closely enough to catch unusual activity early. In this case, it was exactly that kind of monitoring, at two customer sites, that let PaperCut reconstruct the attack chain and warn everyone else before it spread further.
If you'd like us to review your exposure to attacks like this one, please get in touch today.