Open-source tool exposes security risks and detection gaps in AI API relays
开源工具揭露AI API中转站安全风险与检测差异
api-relay-audit audits AI API relay risks with verifiable three-state decisions and transparent logs, covering AC-1 tool-call rewriting, AC-2 error-response leakage, and context truncation, while the author has published the methodology, comparison results, quick-reference table, and the open-source tool.
Why it matters: The open-source api-relay-audit tool splits relay-station behavior into three reproducible checks: rewriting tool calls, leaking model identity through errors, and truncating context, with methods and comparisons published.