Ensuring the security of generative AI applications is more critical than ever.
At Lakera, we are committed to providing comprehensive security solutions to protect GenAI applications against various threats, including prompt injections and PII disclosure.
As part of our robust security suite, we are excited to announce a significant upgrade to Lakera Guard's Content Moderation capabilities.
Whatâs New With Lakera Guardâs Content Moderation
The latest version of Lakera Guard brings several key enhancements.
Accuracy Improvements
We have nearly doubled the accuracy across all moderation categories. This means fewer false positives and negatives, ensuring that harmful content is effectively flagged and managed.
Latency Reduction
By optimizing our backend processes, we have reduced latency by 50x for long prompts. This improvement allows for faster content analysis and real-time moderation.
Profanity Detection
We have introduced a new profanity detector within the moderation endpoint. This feature flags any words or phrases considered offensive or inappropriate, including cursing, sexually offensive terms, religious blasphemies, and abusive phrases directed against ethnic groups or minorities.
Broadened Scope
Hate Speech
The hate speech detector now flags any content that expresses, incites, or promotes harassing language directed at any target, including protected groups. This includes content involving violence or harm toward any target and discrimination through the use of slurs.
Sexually Explicit Content
The detector now also flags sex education, medical, or wellness-related content. It will flag sexual content regardless of whether it is vulgar.
Ready for Enterprise Deployment
With these upgrades, Lakera Guard extends its enterprise-grade capabilities to protect GenAI applications in production. Our updated content moderation capabilities provide the unique mix of high accuracy and low latency required for GenAI use cases.
These improvements are available now to all Lakera SaaS and self-hosted customers. Comprehensive documentation and support are available to help with the transition.
Lakera Guardâs new enterprise-grade content moderation capabilities represent a significant leap forward in our mission to provide the strongest defenses for enterprise GenAI infrastructure.
â
For more information, visit our documentation or contact our support team at support@lakera.ai.
