New Cyber-OSINT model released (twitter.com)

A new Cyber-OSINT model has been released that offers significant capabilities for local deployment, enabling cybersecurity professionals to run advanced operations without relying on external servers. This model utilizes a Mixture of Experts (MoE) architecture with a total of 26 billion parameters, allowing for 4 billion of these to be activated concurrently, and is trained on a vast dataset of 6,500 OSINT (Open Source Intelligence) and CTI (Cyber Threat Intelligence) instructions. Key features include threat-actor attribution, geolocation, and the ability to utilize indicators of compromise (IoCs) effectively, which are crucial for enhancing investigative methodologies in cybersecurity. This release is particularly noteworthy for the AI/ML community as it signifies a more accessible solution for both offensive and defensive cybersecurity strategies. Unlike most existing models that rely solely on system prompts, this new model is fine-tuned for specific cyber tasks, featuring extensive context capabilities of 262,000. Moreover, the model can run on an 8GB GPU, making advanced cybersecurity tools feasible for a wider range of practitioners. This local deployment not only increases security and reduces latency but also aligns with the ongoing trend of empowering security teams with robust, self-sufficient tools in an ever-evolving threat landscape.