AlanH0 @alan_h0 - Twitter Profile

alan_h0 retweeted

11 months ago

Other than hacking and testing, if we target to make higher impact and attempt to make the test and results are more scalable and promising, we cannot avoid to work on engineering and software development. We have developed an enterprise-grade software called VXAI Suite, which is a LLM Tester represents a breakthrough in AI security testing through AI-Testing-AI methodology. This platform leverages local Large Language Models to comprehensively evaluate, challenge, and validate AI systems while maintaining complete data sovereignty and privacy. Core Innovation: Local AI Testing AI LLM Tester introduces AI-native security testing where local LLMs systematically evaluate other AI systems: - Local LLM Intelligence: Gemma2:4b and Yi:6b models for advanced reasoning - Complete Data Sovereignty: Zero external dependencies, all processing local - RAG-Enhanced Knowledge: Domain-specific intelligence for contextual testing - Advanced Mutation Engine: Seven distinct strategies for comprehensive vulnerability discovery - Reinforcement Testing: The goal is to achieve a particular state or outcome through an iterative process of taking actions and receiving feedback. Ensure our testing agent rewards or penalties based on its actions and outcome from target chatbot. - Enterprise Architecture: Project-based isolation with professional reporting We will publish a blog soon for more details, and currently we have business use cases in AI red team and testing already. Please stay tuned. Team: - Project lead, Software architect and lead developer: @alan_h0 - Engine and Algorithm Researcher: @darkfloyd1014 @02E774 @wwkenwong #VXRL @vxresearch

darkfloyd1014's tweet photo. Other than hacking and testing, if we target to make higher impact and attempt to make the test and results are more scalable and promising, we cannot avoid to work on engineering and software development.

We have developed an enterprise-grade software called VXAI Suite, which is a LLM Tester represents a breakthrough in AI security testing through AI-Testing-AI methodology. This platform leverages local Large Language Models to comprehensively evaluate, challenge, and validate AI systems while maintaining complete data sovereignty and privacy.

Core Innovation:
Local AI Testing AI
LLM Tester introduces AI-native security testing where local LLMs systematically evaluate other AI systems:

- Local LLM Intelligence: Gemma2:4b and Yi:6b models for advanced reasoning

- Complete Data Sovereignty: Zero external dependencies, all processing local

- RAG-Enhanced Knowledge: Domain-specific intelligence for contextual testing

- Advanced Mutation Engine: Seven distinct strategies for comprehensive vulnerability discovery

- Reinforcement Testing: The goal is to achieve a particular state or outcome through an iterative process of taking actions and receiving feedback. Ensure our testing agent rewards or penalties based on its actions and outcome from target chatbot.

- Enterprise Architecture: Project-based isolation with professional reporting

We will publish a blog soon for more details, and currently we have business use cases in AI red team and testing already. Please stay tuned.

Team:
- Project lead, Software architect and lead developer: @alan_h0
- Engine and Algorithm Researcher: @darkfloyd1014 @02E774 @wwkenwong

#VXRL
@vxresearch

0

5

3

2

983

alan_h0 retweeted

VXCON

@vxresearch

over 1 year ago

We just published a blog titled "Enhancing LLM Code Generation with RAG and AST-based Chunking", authored by Ken Wong. https://t.co/ZZbjvsiJDp

1

7

2

1K

AlanH0 @alan_h0

over 1 year ago

https://t.co/jSXfRgMd1B

OffSec

@offsectraining

over 1 year ago

We're thrilled to sponsor @vxresearch! Visit our booth on November 16th to learn more about our training and how we can help advance your cybersecurity skills. If you have yet to grab a VXCON pass, use code "VXCONOffsec2024" at checkout to claim a 10% discount on your VXCON 2024 tickets: https://t.co/AEYYF78Tg1

0

9

1

2

5K

0

45