We’ve developed an intent modeling system using embedding-based approaches that leverage large language models like BERT and GPT.
Intent Modeling
Using LLMs to produce embeddings that represent semantic meaning of a word or phrase in order to categorize text.
What is the item?
The system goes beyond simple phrase matching, using semantic clustering to understand intent in varied phrasing, allowing it to recognize new and evolving attack patterns.
Why is it helpful to our customers?
This innovation enables rapid, accurate detection of malicious intent, even when attackers use novel language variations. It eliminates the need to enumerate variations upfront, making our system more automated and resilient.
Why is it interesting?
The integration of advanced language models allows Abnormal Security to adapt quickly to unseen attack patterns, a required trait in an adversarial environment.
The flexibility of this intent modeling approach allows us to easily add new intents for the constant new stream of attack types.