Lasso Security's guardrail skips the GPU

Lasso Security announced a $30 million funding round on September 2 alongside LEAP, a guardrail engine the company describes as transformer-free: it screens AI traffic for policy violations on standard CPUs, with no GPU required, returning a verdict in under five milliseconds at what Lasso says is thousands of times the throughput of existing guardrails. ClearSky led the round, joined by Entrée Capital, iAngels, Singtel Innov8, Mindset, and Swish Data.

The platform pairs LEAP with a second engine, RAPID, a self-hosted model that handles the smaller share of decisions requiring actual judgment against a plain-language policy, at what Lasso says is a fraction of what routing the same call through a commercial model provider would cost. A routing layer decides which engine sees each request, so the bulk of traffic never touches a graphics processor at all. Lasso says revenue grew more than 500% over the past 12 months, with customers including the US Department of Homeland Security, Kaufman Rossin, eToro, and Fiverr.

The pitch targets a real constraint: running a guardrail on every request, rather than sampling traffic, gets expensive fast when each check needs a GPU. A CPU-first detection layer changes that cost math for teams that have been rationing inspection coverage rather than screening everything. It’s a narrower, product-specific move than the platform consolidation happening elsewhere in agent security, like Palo Alto Networks’ acquisition of Console for Cortex, but it’s aimed at the same underlying problem: making agent oversight cheap enough to run on every action instead of a sample of them, the same design goal behind JetStream’s newly announced per-action authorization engine.