rag-inject-guard
Copyright 2026 Fevzi Ege Yurtsevenler

This product is licensed under the Apache License, Version 2.0 (the "License");
you may not use this software except in compliance with the License. You may
obtain a copy of the License in the accompanying LICENSE file or at:

    http://www.apache.org/licenses/LICENSE-2.0

------------------------------------------------------------------------------
Prior art and acknowledgements
------------------------------------------------------------------------------
This project stands on public work by others. It re-implements no third-party
code; the acknowledgements below credit the ideas and threat framing it builds on:

- Simon Willison - the "lethal trifecta" framing for exfiltration risk
  (private data + untrusted content + external communication).
- Meta - LlamaFirewall (PromptGuard 2, AlignmentCheck), an open-source
  guardrail system for AI agents (arXiv:2505.03574).
- StackOne - Defender, an open-source indirect-prompt-injection detector
  (pattern + lightweight ML classifier).
- OWASP - GenAI / LLM Top 10 (LLM01 Prompt Injection).
- MITRE ATLAS - adversarial ML threat taxonomy.
- The "Trojan Source" research (Boucher & Anderson) on bidirectional-override
  (CVE-2021-42574) and homoglyph (CVE-2021-42694) source-code attacks.

The optional integrations import LangChain (langchain-core) and LlamaIndex
(llama-index-core) only when explicitly used; neither is a required dependency.
