Using large language models to outsmart cyber attackers

Honeypots are a well-known tool in cyber security. They are computers that attract hackers' attention, either to divert attacks from an organization's real systems or to learn more about the methods attackers use.

"We have built expertise on large language models and federated learning over several years. Federated learning is a method that allows multiple stakeholders to collaborate in training models. The honeypot project lets us fuse these areas into practical applications within the security domain," says Dr. Mats Nordlund, Director of AI Labs at AI Sweden. He adds:

"Since 2021, AI and security have been part of the work in our Edge Learning Lab, following recommendations from our advisory board. We are now further accelerating our efforts in this area, under the umbrella AI & Security, as the intersection of AI and security is becoming increasingly critical for many of our partners. This includes new projects aimed at making AI itself more robust and secure, as well as exploring how AI can be leveraged to enhance security in other domains. Our work on adaptive honeypots is one example of the latter."