LLM Vulnerability scanner and guardrails

LLM Vulnerability Scanner and Guardrails provides comprehensive assessment of LLM vulnerabilities and automatic application of optimal defensive techniques to generative AI on LLMs.

The technology addresses the challenge that the number of new attacks against generative AI is increasing and that numerous real-world attack examples are being reported. 
New attack methods are emerging one after another, making it difficult to implement countermeasures due to a shortage of experts.

The solution enables even non-experts to operate generative AI securely.

The LLM Vulnerability Scanner performs a comprehensive investigation of LLM vulnerabilities. It supports over 7,700 vulnerabilities. The scanner assesses potential defensive capabilities against new attacks by evaluating not only simple attacks but also complex ones.

The LLM Guardrails component automatically addresses vulnerabilities. It automatically defends against attacks by applying rules generated based on the investigation results and vulnerability information to the generative AI system. The technology automatically generates check rules for detected vulnerabilities through the integration of scanners and guardrails and by applying LLM guardrails.

The technology was developed in joint research with Ben-Gurion University to address threats from attacks that are difficult to analyse.