Announcing LLM Guard 0.3.3 Release: Laiyer AI Prompt Injection Model ⚡
We're excited to roll out version 0.3.3 of LLM Guard, our latest update that continues to elevate the efficiency and robustness of our toolkit for securing LLMs.
At the core of the update is our newly released proprietary prompt injection model that we’ve open-sourced to the broader LLM Security community and that greatly outperforms the previous state-of-the-art alternatives in the market.
Over the past months, we’ve seen that there has been no concerted effort in OS to release an updated model that is able to detect and secure against novel prompt injection attacks. That’s why we decided to pick up the baton of great work already done by other folks like Jasper Schwenzow at Deepset and developed our very own model. We believe any company should be able to instill security in any LLM they wish to use or embed into their day-to-day activities or products. That is exactly why we decided to open-source it for the broader community.
Besides that, we’ve also noticed a challenge for security and product professionals to evaluate LLM security alternatives in the market. For example, what is the difference between different prompt injection detection models and how do they actually perform in action - i.e. marketing vs. reality? We have something unique in the works that will allow anyone to get a better understanding of this. Stay tuned and subscribed if you’d like to receive this.
Moving over to our current release, here’s a glimpse of what’s new:
🌟 Proprietary Prompt Injection Detection Model: Introducing 'laiyer/deberta-v3-base-prompt-injection', our new proprietary model designed to offer advanced detection capabilities for prompt injection threats, significantly bolstering the security of your LLM deployments.
📊 New Benchmarks on Azure Instances: We have introduced updated benchmarks specifically tailored for Azure instances, ensuring you get the most accurate performance data for cloud-based deployments.
🔧 Upgraded json_repair Library: We have upgraded the json_repair library to enhance stability and efficiency in handling JSON data. Big thanks to the suggestion by the author of the JSON library, Stefano Baccianella.
Stay Updated and Secure
To learn more about these updates and how they can benefit your LLM security strategy, we invite you to visit our comprehensive documentation. Dive into the specifics of each new feature and upgrade, and see how LLM Guard continues to be the leading toolkit for LLM security.
Stay in Touch
To keep abreast of our latest updates and releases, follow us on:
We're deeply thankful to our vibrant open-source community for their ongoing contributions and dedication. To get involved or explore the latest developments, visit our Github page.
Thank you for being a part of this exciting journey with LLM Guard! 🌟