InjectionShield is a lightweight, deterministic Python module that acts as a gate in front of AI agents to block prompt injection attacks. It uses rule-based detection on untrusted inputs such as emails or tool outputs, produces measurable error rates on labeled test sets, and requires no network, API, or ML model. The accompanying guide walks users through building, testing, and evaluating the filter with safe and malicious prompt corpora.
InjectionShield is a LLM eval & observability project. It focuses on preventing untrusted input from injecting malicious instructions into AI agents without relying on cloud services or machine learning. InjectionShield is an open-source project aimed at AI security researchers and developers. InjectionShield is open source under the Open Source license. It runs on the web and the command line, and it can be self-hosted.
InjectionShield first shipped in 2025. Key capabilities include Prompt Injection Filter, Deterministic Rules, and Test Evaluation.
Summary written by a language model from the project’s public pages.
What PulseGate has recorded for this listing
Closest matches by what these projects do