Lasso Security, the AI security company, announces LEAP, a new transformer-free class of AI guardrail that delivers top-tier detection accuracy on ordinary CPUs -- no GPU required -- at under five milliseconds per decision and thousands of times the throughput of existing guardrails.
LEAP, which is already in production at global enterprises and the U.S. federal government, is the first of two engines in the Lasso platform (the other is RAPID) built on a simple idea: companies should not need access to a supercomputer to determine whether a single AI request is safe.
"Two years ago, the question was whether enterprises would put AI in front of their customers. That question is settled," said Elad Schulman, Co-Founder and CEO of Lasso Security. "What they need now is security that keeps up with agents that act autonomously and at scale. This is what LEAP was built for -- inspecting not only what agents say, but also what they do, at enterprise scale and cost-effectively."
Until now, the industry's reflex has been to put another large model in front of the AI, with a reasoning LLM inspecting every request. That model needs accelerated hardware (a GPU) on every request, for every user, and the bill arrives whether the request was dangerous or not. Most enterprises respond the only way they can: they inspect part of their AI traffic and accept the risk on the rest.
"The people building security at the largest scale have started saying out loud what the architecture makes obvious: reasoning models are expensive and built for deliberation, not for the continuous, real-time and low latency monitoring that real AI security demands," said Ophir Dror, Co-Founder and CPO of Lasso Security.
LEAP removes that constraint. It is transformer-free and runs with zero GPU footprint, delivering leading detection quality at a fraction of the cost, with minimal latency and no context-window restrictions. In Lasso's head-to-head benchmarks, no other guardrail matches its combination of accuracy and throughput. It sits alone in the top-right of the field, as accurate as models that need dedicated hardware and hundreds to thousands of times faster. Because there is no specialized hardware to reserve, LEAP deploys anywhere the customer's data has to stay, be it the customer's own cloud, regulated environments or air-gapped networks.
"The concept of a guardian agent or agents overseeing other agents is great, unless you are the one paying the bill," said Dror. "In real world deployments with millions of prompts and actions, this idea is just not feasible."
Lasso does not pretend the hard cases go away. Its platform pairs LEAP with RAPID, both of which are patent pending. RAPID is a self-hosted LLM-as-a-judge that handles the small fraction of decisions requiring complex, plain-language policy judgment, at 100-200x lower cost than calling a cloud LLM API. A tiered routing layer sends the overwhelming majority of traffic to LEAP for an inline decision in milliseconds, and escalates only what genuinely needs deeper reasoning to RAPID where it belongs, not on every request. The GPU does not disappear. Lasso just stops spending it on every request.
Lasso reached this point by attacking AI systems before defending them. Its offensive team runs automated red teaming against customers' AI applications and agents, from single-prompt attacks to autonomous multi-turn adversaries, and feeds what it finds straight back to the product. In one engagement with a global healthcare provider, Lasso's red team extracted all of the patients records, and was able to alter prescriptions provided to patients.
Lasso secures AI for global enterprises across financial services, insurance, hospitality and automotive, and the U.S. federal government, inside customers' own environments, protecting tens of thousands of agents and billions of requests and actions per month.
In conjunction with the technology news, Lasso is announcing a $30 million funding round led by ClearSky, with participation from Entrée Capital, iAngels, Singtel Innov8, Mindset and Swish Data. Lasso plans to use the funding to expand engineering, deepen its federal and regulated-industry work, and grow go-to-market in North America and Europe.
About Lasso Security
Lasso Security is an AI security platform providing observability, governance, and real-time defense across the entire AI ecosystem, securing the AI models enterprises use, the agents they build, and the applications they ship. Founded in 2023, Lasso is SOC 2 Type II certified.
