Show HN: Keep your credentials out of the AI agents you use and build

Category: security

Tags: security, ai-agents, credential-leak-detection, llm-observability, canary-tokens

Score: 7.8/10 (Innovation: 8, Technical: 7, Documentation: 9, Utility: 7)

Weir is a security tool that detects when AI agents leak credentials by planting unique fake keys in the agent's environment, running its flows, and checking every output channel (spans, logs, prompts, network) for those exact values. It also integrates with Claude Code to monitor and redact real credentials during sessions. Notably, it was used to discover a real vulnerability in MLflow.

Target audience: AI/ML engineers, security engineers, backend developers building agentic systems

Repository: https://github.com/IdoGol24/weir · Python · Apache-2.0 · 18 stars

View on Hacker News