AI Governance & Security Vendor Wiki

6 min read

Lasso Security

Primary category: dlp. Spans ai-runtime-security, agent-runtime-security, and ai-access-governance.

One-liner — An Israeli end-to-end GenAI security platform that watches every LLM and agent touchpoint (employee chatbots, built apps, MCP tools) to stop sensitive-data leakage, prompt injection, and shadow-AI sprawl in real time.

What it does

Lasso protects the full lifecycle of LLM/GenAI use inside an enterprise. It does three jobs that the wiki splits across categories:

  • Shadow-AI discovery — its Discover module maps the AI models, agents, and apps in use across the org, including unapproved (“shadow”) tools employees adopt on their own.
  • Context-based data protection / DLP — inspects prompts and model responses for sensitive-data disclosure and policy violations, and can block before data egresses to an LLM.
  • Runtime guardrails — detection models (trained on a large, frequently-updated corpus of prompt-injection and jailbreak attempts) screen inputs and outputs for prompt injection, jailbreaks, malicious code generation, model/data exfiltration.
  • Agent / MCP security — an open-source MCP Gateway plus an “Intent Security Framework” that baselines what an agent should do, which tools it should call, and what data it should touch, then flags deviations.

The pitch to a CTO: one platform to see and govern both human AI use (employees pasting into ChatGPT) and built AI (your own LLM apps and agents), rather than buying a separate shadow-AI tool, an AI firewall, and an agent-security tool.

Where it sits in the stack

Primary home is dlp (Data layer) because the headline job is keeping sensitive data out of prompts and responses. But Lasso is genuinely cross-layer:

Risk coverage: it screens untrusted input (prompt injection/jailbreak detection), guards sensitive data (DLP on prompts/responses), and controls egress (blocking leakage to external LLMs and via MCP tool calls). It lives at the boundary between the green zone (internal data/agents) and the red zone (external models and tools).

Deployment & architecture

Multiple insertion points rather than a single chokepoint:

  • Browser extension — “Lasso for Employees,” to see and control AI use in the browser (shadow-AI, employee chatbot guardrails).
  • Inline / reverse proxy + API gateway + SDK — “Lasso for Applications,” to wrap your own LLM apps; enforce policy inline at the proxy, API, or AI-gateway layer.
  • Open-source MCP Gateway — a proxy/orchestrator for Model Context Protocol traffic (on GitHub, Smithery, MCP Servers), for agentic workflows.
  • SaaS control plane with dashboards/observability.

Integrations include AI gateways — e.g. Lasso ships as a guardrail in portkey’s gateway. Vendor performance claims (marketing): sub-50ms decisions, 99.8% detection accuracy.

Positioning & differentiators

  • Breadth is the story: most neighbors pick one lane (pure DLP, pure AI firewall, pure shadow-AI). Lasso spans employee AI + built AI + agents/MCP under one platform.
  • MCP/agent focus early — it shipped an open-source MCP security gateway (Apr 2025) and markets an agent “Intent Security Framework,” positioning it toward agentic deployments, not just chatbot DLP.
  • Origin in LLM threat research — leans on a large prompt-injection/jailbreak dataset as the basis for its detection models.

Nearest neighbors: prompt-security (very similar end-to-end GenAI scope; acquired by SentinelOne), witnessai (runtime + AI access governance), harmonic-security and cyberhaven (shadow-AI / DLP for AI), nightfall-ai and mind (data-centric DLP), pillar-security and hiddenlayer (AI runtime), zenity (agent runtime).

Ownership, funding & M&A

  • Independent, VC-backed. No M&A — Lasso has not been acquired (and the seed registry carried no acquisition flag). Confidence: high that it remains independent.
  • Seed: $6M, announced 2023-11-20, led by Entrée Capital with Samsung Next (verified, vendor + SecurityWeek). Confidence high.
  • Later funding: aggregators (Crunchbase/PitchBook/Tracxn) report additional SAFE/round activity in 2024-2025 (e.g. a ~$10M SAFE in mid-2025 reportedly including CyberArk Ventures and Singtel Innov8, and a Series A reported late 2025; ~$28M total cited). These are not confirmed against a primary source and figures conflict between aggregators — treat post-seed totals as unverified.

CTO / hedge-fund lens

  • Day-1-ish, but as a consolidation play. A fund’s Day-1 AI controls are “stop staff leaking data into ChatGPT” (shadow-AI + DLP) and “guardrail any internal LLM app.” Lasso covers both, plus agent/MCP for when you go agentic (Day-2). For a small fund that wants one vendor instead of three, that breadth is attractive.
  • Model-risk / SR 11-7: Lasso is a control/monitoring layer, not a model-governance platform — it helps with the data-leakage and prompt-abuse side, not model validation or documentation. Pair with an ai-governance-platform for SR 11-7 obligations.
  • Fit: medium. Strong for a shop standing up employee AI + a few internal LLM/agent apps and wanting a single pane. Less compelling if you already own DLP via netskope/microsoft-purview or an AI firewall via prisma-airs. Early-stage vendor — diligence the roadmap, support, and SOC 2 posture.

Competitors / alternatives

prompt-security, witnessai, harmonic-security, cyberhaven, nightfall-ai, mind, pillar-security, hiddenlayer, prisma-airs, zenity.

Open questions / to verify

  • Post-seed funding: confirm the 2024-2025 round(s), exact amounts, investors, and total raised against a primary source (current data is aggregator-only and conflicting).
  • Confirm the four co-founders (Elad Schulman CEO; Lior Ziv, Ophir Dror, Yuval Abadi per press) — names not all in vendor primary docs.
  • Customer base / scale and compliance certifications (SOC 2, ISO) — not established.

Sources

History

  • [2026-06-28] Stub created from seed registry.
  • [2026-06-28] Researched; established founding (2023), HQ (Tel Aviv), independent ownership, $6M seed (Entrée Capital/Samsung Next, 2023-11-20, verified). Confirmed cross-layer scope (DLP + AI runtime + agent/MCP + shadow-AI) and deployment (browser extension, inline/reverse proxy, API gateway, SDK, open-source MCP Gateway). No M&A. Post-seed funding reported by aggregators but left unverified. Set status: researched; sources_count 4.

Graph View

Backlinks