Lasso Security

Lasso is a runtime security and governance platform for LLM and agent traffic, not a single tool. The product spans five stages: discovery of AI agents and MCP connections across an org (an "AI-BOM"), posture assessment against frameworks like NIST and OWASP, automated red-teaming of agents with a vendor-claimed library of 3,000+ attack techniques, inline policy enforcement at the proxy/gateway layer, and detection/response for live attacks (prompt injection, tool poisoning, data exfiltration). It sits as a proxy in front of LLM calls and MCP servers rather than as a code-scanning or model-training tool.

Lasso publishes an open-source MCP Gateway (github.com/lasso-security/mcp-gateway, MIT license, 385 stars, 38 forks) that any team can pip install mcp-gateway and run today, without a sales call. It's a genuine plugin-based proxy for MCP servers with three built-in guardrail plugins — basic secret-masking, Microsoft Presidio PII detection, and a lasso plugin that calls Lasso's paid API for the fuller feature set (prompt-injection detection, custom policy, harmful-content filtering). This is unusual for the category: most AI-security vendors sell nothing you can evaluate without a demo call. The free-tier plugins alone (secret masking + PII) are a real, no-cost starting point for MCP hygiene.

Lasso's research team has produced findings that were picked up by outlets outside its own blog: an investigation that found over 1,500 exposed HuggingFace API tokens (including from Meta, Microsoft, Google, VMware), covered by The Decoder (Dec 2023); and earlier research on "AI package hallucination" — LLMs recommending non-existent open-source packages that attackers could register and weaponize, including a documented case where a hallucinated PyPI package the researcher registered as a honeypot got 30,000+ downloads in three months. Both are genuine, methodologically described findings, not just marketing claims.

Lasso was founded in 2023 by CEO Elad Schulman (previously co-founded Segasec, a phishing-protection company acquired by Mimecast for $50M in 2020) and CTO Lior Ziv. Its only publicly disclosed funding is a $6M seed round (Nov 2023) led by Entrée Capital with participation from Samsung Next — SiliconAngle put total outside funding at "just over $7.5M" at that time. No Series A or later round is publicly disclosed as of this check. StartupHub.ai estimates ~50 employees (a third-party estimate, not vendor-confirmed). The company lists enterprise logos including the U.S. Department of Homeland Security, Optibus, Fiverr, and eToro on its site, and says it was named a Gartner Cool Vendor in AI Security for 2024 — a claim repeated in multiple PR write-ups (GlobalSecurityMag, Swish Data) but not independently confirmed against a Gartner primary source in this check.

No public pricing anywhere on the site — the entire commercial motion is "Book a Demo." There is no self-serve tier, published "from $X" figure, or free trial mentioned outside the open-source MCP Gateway. A real Information Security Policy PDF is published (last approved by the company's COO, references SOC 2/ISO 27001/PCI DSS/GDPR as compliance targets the company says it works toward) — but this is a policy statement, not a certificate; no SOC 2 report, trust portal, or third-party audit attestation is linked from the site.

The MCP Gateway repo was created April 2025 and last had a commit pushed January 22, 2026 — over seven months before this check, despite the repo remaining open and actively discussed in third-party MCP-gateway comparison posts (MintMCP, AimMultiple, Zuplo). The matching PyPI package (mcp-gateway) last released v1.2.1 the same day. 12 open issues, no recent releases since. This isn't abandoned, but it's not under active weekly development either.