Invariant

Invariant Labs builds security and observability tooling for AI agents. It was acquired by Snyk, an announcement shown prominently on its site, and describes its mission as helping agent builders create secure, reliable and robust products. The team comes from academic research in robust, secure and private AI, with advisors from ETH Zurich, and previously created the LMQL language and the LVE repository of language model vulnerabilities. The product family has three parts. Explorer lets developers inspect and observe agent behavior, which is useful for debugging agent traces. Guardrails is a contextual security layer that sits around an AI agent to enforce rules on what it can do. MCP Scan is a security scanning tool for Model Context Protocol servers, intended to help users check the tools their agents connect to before trusting them. The tools are aimed at developers and security teams building or deploying agents that call tools and external services, and they are meant for testing and protecting the buyer's own systems. The site does not publish plan prices. Visitors can try the products from the home page or subscribe to a newsletter.