Lasso Security, a Tel Aviv and New York-based startup that screens the actions of AI agents for security risks, said Tuesday it has raised $30 million in new funding led by ClearSky Advisors, with existing investor Entrée Capital increasing its stake alongside iAngels, Singtel Innov8, Mindset and Swish Data. The round brings the company's total funding to more than $37 million since it was founded in 2023.
The raise accompanies the launch of LEAP, a "transformer-free" guardrail engine that Lasso says can evaluate whether an AI agent's request is safe to execute using ordinary server processors, without routing traffic through a GPU. The company says LEAP returns a verdict in under five milliseconds and can handle far higher request volumes than guardrails built on large language models, which typically require GPU inference and add latency to every step an agent takes.
LEAP is paired with a second engine, RAPID, a self-hosted model reserved for more nuanced policy calls; a routing layer sends most traffic through the faster, cheaper CPU path and escalates only ambiguous cases to the heavier model. Lasso co-founder and Chief Product Officer Ophir Dror argued that a common alternative — using a second AI agent to supervise a first one — does not hold up at production scale.
The concept of a guardian agent... is just not feasible.
Ophir Dror, co-founder and chief product officer, Lasso Security
Lasso, co-founded by chief executive Elad Schulman and Dror, counts BMW, eToro, Fiverr and the U.S. Department of Homeland Security among its customers. The company says its platform now covers tens of thousands of AI agents processing billions of requests and actions a month, with revenue up more than 500% over the past year.
The funding lands as enterprises increasingly deploy autonomous AI agents that take actions — sending emails, modifying records, calling outside tools — rather than simply answering questions, raising the stakes of a single unchecked bad decision. Guardrail and AI-governance startups have drawn a wave of venture interest this year as a result, competing on how cheaply and quickly they can screen agent behavior without slowing it down. Schulman said the debate over whether enterprises should deploy AI agents at scale is now settled; in his telling, the open question is how to make oversight of those agents cheap enough to run on every request rather than a sample of them.
