api-relay-audit

To audit an AI API relay for prompt injection and data leakage, run a local end-to-end scan that sends deterministic probes to detect hidden prompt tampering, identity hijacking, and error-response credential leaks, yielding a structured Markdown risk report.

API relay prompt tampering occurs when a third-party proxy alters requests or responses by injecting hidden instructions, truncating context, or leaking prompts. It compromises LLM security by enabling instruction overrides and unauthorized data exfiltration.

Yes, you can check SSE stream integrity for Anthropic-compatible endpoints by validating stream invariants against deterministic probe sets. This detects context truncation and stream manipulation anomalies introduced by the relay.

Yes, this API relay audit detects Web3 wallet-safety risks and tool-call substitution by applying profile-gated checks for package substitution and malicious prompt injection designed to compromise agent signing or transfer operations.

You do not need complex Python dependencies to run a local security audit on an OpenAI API proxy; you can execute a single standalone script using curl-only transport, though a modular Python implementation with httpx is also supported.

The best way to compare multiple AI proxy providers for supply chain safety is to run deterministic audit probes against each endpoint, evaluating risk scores for prompt leakage, tool-call substitution, and infrastructure fingerprinting to generate comparative reports.