Your Agent Is Mine: Measuring Malicious Intermediary Attacks on the LLM Supply Chain

The project is a academic research artifact associated with a recent Arxiv paper. While it addresses a critical and overlooked vulnerability—the lack of end-to-end cryptographic integrity in LLM API routers (like OpenRouter, LiteLLM, or internal enterprise proxies)—it is currently a reference implementation for measuring risk rather than a persistent security tool or platform. With 0 stars and 6 forks, it is in the early dissemination phase, likely among peer reviewers or fellow researchers. Its defensibility is low because it is designed to highlight a systemic flaw that, once publicized, will likely be addressed by protocol changes (e.g., mTLS, signed payloads) or integrated into professional security scanners like Giskard or Protect AI's suite. Frontier labs have a low risk of building this as a product, but they are highly likely to implement the security recommendations (cryptographic signatures) that would render this measurement tool obsolete. The primary value is the 'breakthrough' realization of the threat model rather than the code itself.