Show HN: AgentShield – Stop AI agents from spending money unsupervised

Category: security

Tags: ai-agents, spending-control, payment-firewall, python-sdk, stablecoin

Score: 6.8/10 (Innovation: 6, Technical: 7, Documentation: 8, Utility: 6)

AgentShield provides a spending firewall for AI agents, intercepting payment requests and running quantitative, policy, semantic, and goal-drift checks to approve, flag for human review, or block transactions. It's interesting because it addresses a real and growing need for guardrails around autonomous agent spending, with thoughtful design decisions like atomic budget reservations, HMAC signing, and fail-safe LLM fallbacks.

Target audience: backend devs, AI engineers, security engineers

Repository: https://agentshieldv2-dashboard-production.up.railway.app/ · Python · MIT · 4 stars

View on Hacker News