In an environment where cyber threats are becoming increasingly complex, understanding, anticipating and countering attacks requires cross-sectoral skills and advanced technologies. This is why AI4Cyber studio, the new applied research space of Tinexta Defencededicated to the research and application of Artificial Intelligence in cybersecurity.
A multidisciplinary team that addresses the challenges of the cyber landscape, it supports clients in adopting AI by sharing research results with the scientific and technical community.
The entire lifecycle of AI systems - from data collection to deployment - is managed with Machine Learning, Deep Learning and LLM technologies, respecting the principles of ethics, transparency and privacy protection.
The first in-depth study is devoted to an experimental project integrating language models in high-interaction deception technologies.
The study presents a framework for the creation of a high interaction honeypot (HHI) using LLM locallycapable of simulating a Linux terminal and responding credibly to attackers' commands by tracking their actions. The architecture consists of three modules: Terminal Protocol Proxy, Prompt Manager and LLM Engine.
To handle the contextual limits of the models, an algorithm of pruning and a classification system based on an 'F' value, which evaluates the impact of received commands.
The approach has been tested with several open-source models (Mistral, Gemma2, Phi4, Llama3.1, Qwen2.5) and advanced versions such as ChatGPT and Claude AIconfirming its validity.
If you wish to learn more, here is the link to our studio complete.
In addition, you can subscribe to the specific mailing list Cyber Studios by Tinexta Defenceto receive updates on upcoming research:
