How to apply AI to Email Security
Artificial Intelligence (AI) has undergone a remarkable transformation over the past few decades, evolving from simple rule-based programs to sophisticated, self-learning systems. In cybersecurity, this evolution has enabled more advanced threat detection, faster incident response, and even predictive defense mechanisms.
‘Generation 1’, rule based, or ‘symbolic AI’ AI had a direct and immediate impact to solutions such as Anti-Virus, Intrusion Detection Systems and Email filters when dealing with known pattern based spam. It was not until the second generation of Machine Learning and Predictive AI that Email security solutions started to use AI in any seriousness, but why then are email threats still the most ‘unsolved’ problem in the world of Cybersecurity?
“The performance of your AI model will only be as good as the quality and relevance of the Data it is trained on.”
— Frank Zhang, Chief Data Scientist, Sabiki Security.
Sabiki Email Security provides a capability and workflow that allows administrators to easily train their own tenant specific AI model for Email Security.
The problem with applying AI models to Email Security is that vendors pre-train their models with email that is not yours. While this provides a ‘good start’ at increasing capture rates, Sabiki allows for a fundamental truth of Artificial Intelligence to be put into practice.
Good Data in, Good Data out.
Passive and Active learning
User initiated Training
In User approved mode, Sabiki initiates live training cycles based on user activity within their mail client. Ideal for power users.
Admin Controlled AI training
For a broader and more controlled deployment. The Email administrator selects which emails are trained by the model as ‘Good’ and ‘Bad’ instantly updating and applying the learning into production.
Data Science, without the Data Scientist
