Resk-Security launches open-source LLM firewall to block harmful tokens

Resk-Security has released reskSecure, a new open-source tool designed to enhance LLM security by filtering at the logits level before token generation. This preventive approach aims to block harmful tokens from being sampled, unlike traditional methods that filter output after generation. The tool utilizes a capability bitmask system to penalize or block specific tokens based on user-defined policies, offering modes like HARD (impossible to generate) and BIAS (configurable penalty). reskSecure can be integrated into Python applications and supports hot-reloading of policy files for continuous security. AI

IMPACT Enhances LLM security by preventing harmful token generation, potentially reducing risks in AI applications.

RANK_REASON The cluster describes the release of a new open-source tool for LLM security.

Read on dev.to — LLM tag →

• GitHub
• Python
• Python Package Index
• reskSecure
• Resk-Security

• product
• safety

AI-generated summary · Google Gemini · from 1 sources. How we write summaries →