Invariant Labs
Invariant Labs is an ETH Zurich spin-off and AI security research lab focused on making AI agents robust, reliable and secure. It builds tools for observing agent behavior, enforcing contextual security policies and scanning Model Context Protocol (MCP) servers for vulnerabilities.
At a Glance
- AI and agent developers
- AI security engineers and security teams
- Platform and infrastructure teams operating enterprise agents
- Organizations deploying LLM, MCP and autonomous-agent applications
- +1 more
AI Tools by Invariant Labs
(1)
Invariant Gateway
Open Source LLM Proxy for Agents
Discussions
No discussions yet
Be the first to start a discussion about Invariant Labs
Latest News
Snyk acquires Invariant Labs to accelerate agentic-AI security innovation
GitHub MCP Exploited: Accessing private repositories via MCP
Invariant Research's AgentDojo wins first prize in the Center for AI Safety SafeBench competition
Invariant MCP-Scan partnership with Smithery
Products & Services
An open-source, transparent contextual guardrailing layer deployed at the LLM/MCP boundary. It evaluates expressive deterministic rules over agent traces and can block unsafe tool calls, data flows and content.
An open-source scanner for MCP installations and servers that analyzes tool descriptions for tool poisoning, MCP rug pulls, cross-origin/tool-shadowing escalation and prompt injection; it also supports tool pinning and inspection.
An observability and visualization tool for browsing, debugging, annotating, testing and sharing AI-agent traces, including a public benchmark registry.
A lightweight, zero-configuration LLM/MCP proxy that forwards requests, captures runtime traces and supports security monitoring and guardrail enforcement without invasive agent-code changes.
Market Position
Invariant Labs positioned itself around the combination of agent observability, contextual runtime guardrails and MCP-specific security scanning. Its differentiator was enforcing organization-specific rules over context and data flows—not only filtering isolated prompts—while keeping the tooling transparent and open source. Following the June 2025 acquisition, these capabilities are being incorporated into Snyk's AI Trust Platform and Snyk Labs.
Leadership
Founders
Marc Fischer
PhD researcher and former doctoral student in ETH Zurich's Secure, Reliable and Intelligent Systems (SRI) Lab; CEO and co-founder. His work focuses on securing and breaking agentic AI systems.
Mislav Balunović
PhD researcher and former doctoral student in ETH Zurich's SRI Lab; former CTO and co-founder. He was a long-time collaborator with Martin Vechev on AI-security research.
Luca Beurer-Kellner
ETH Zurich AI-security researcher and former SRI Lab doctoral student; Chief Technology Officer/Chief Scientist and co-founder. He has worked on LLM security, programming interfaces and agent security.
Martin Vechev
Professor of Computer Science at ETH Zurich and Head of its Secure, Reliable and Intelligent Systems Lab; co-founder and advisor. His lab is known for research in robust, secure and private AI and for producing the LMQL and LVE projects.
Florian Tramèr
Professor of Computer Science at ETH Zurich and researcher in secure/private AI; co-founder and advisor. He has collaborated with Vechev's group on robust and secure AI research.
Executive Team
Marc Fischer
Chief Executive Officer and co-founder
PhD researcher and former ETH Zurich SRI Lab doctoral student focused on agentic-AI security.
Luca Beurer-Kellner
Chief Technology Officer and co-founder
ETH Zurich AI-security researcher and former SRI Lab doctoral student working on LLM security, programming interfaces and agent systems.
Board of Directors
Founding Story
Invariant Labs was spun out of research conducted by its founding team at ETH Zurich over several years. The founders saw autonomous, connected AI agents as a new class of software with security and reliability risks that traditional controls do not adequately address, and formed the company to turn their research into tools that safeguard real-world AI deployments.
Business Model
Revenue Model
The public materials describe open-source tools and free self-serve entry points, alongside custom dedicated/private deployments and organization-level deployments available by contacting Invariant. A public usage or subscription price is not stated.
Target Markets
- AI and agent developers
- AI security engineers and security teams
- Platform and infrastructure teams operating enterprise agents
- Organizations deploying LLM, MCP and autonomous-agent applications
- Researchers and developers benchmarking or testing agent systems
- Securing LLM and MCP-powered agents against prompt injection and tool poisoning
- Preventing sensitive-data and PII leakage through tools, email and other external sinks
- Observing, debugging and auditing autonomous agent behavior
- Testing agent capabilities, reliability and regressions before production
- Scanning and monitoring MCP server/tool ecosystems
- Enterprise security and compliance monitoring across multiple agent applications