As we started deploying self-hosted, multi-tenant AI agents, we realized standard IaC scanners and broad AI governance platforms were missing critical vulnerabilities. General tools simply weren't designed to catch the nuanced, agent-specific isolation defects that can lead to catastrophic cross-tenant data breaches. Initially, we tried adapting existing security scanners, but the approach was too broad. We realized we needed to go deep on a single, critical failure class: tenant isolation. That’s what inspired TenantGuard. We evolved our approach to focus exclusively on real-world exploits, mapping our policies directly to confirmed vulnerabilities rather than theoretical risks. TenantGuard is an open-source (Apache-2.0) CLI and MCP server built specifically to audit multi-tenant AI-agent deployments. Powered by 16 fail-closed OPA/Rego rules, it instantly detects: - Unscoped sandbox mounts - Cross-tenant cron bindings - SSRF-exposed tool URLs - Credential leakage via exec tools It natively outputs to SARIF, JSON, and MCP, so it drops seamlessly into your existing CI/CD pipelines. Repo: https://github.com/RudrenduPaul/... MCP Servers: https://mcpservers.org/servers/r... https://glama.ai/mcp/servers/Rud... PyPI: https://pypi.org/project/tenantg... NPM: https://www.npmjs.com/package/te... We built this for the community and would love your thoughts. What is the biggest security hurdle you’ve faced when deploying AI agents in production?

TenantGuard

Open source scanner for cross-tenant AI-agent security bugs

Open source scanner for cross-tenant AI-agent security bugs