On 2 September 2026, Israeli security startup Lasso Security announced a $30 million funding round led by ClearSky, alongside the launch of LEAP, an AI guardrail product built specifically to run on CPUs rather than the GPUs most AI safety tooling assumes it needs. Entrée Capital, which led Lasso's original $6 million seed round in 2023, increased its position, joined by iAngels, Singtel Innov8, Mindset and Swish Data — bringing total investment in the company past $37 million.
Why "runs on a CPU" is the actual product story
A guardrail, in this context, is the layer that inspects what goes into and out of an AI model — checking prompts and responses for prompt injection, data leakage, policy violations and similar risks — before a request reaches the model or a response reaches the user. Most guardrail tooling to date has run on the same GPU infrastructure as the models it protects, on the assumption that inspecting AI traffic requires AI-scale compute.
LEAP is built to run that inspection on ordinary CPU infrastructure instead. If that holds up under real-world load — worth treating as the vendor's own claim rather than an independently benchmarked one until third parties test it — it changes the unit economics of running safety checks on every request, rather than sampling a subset to keep GPU costs down, which is the compromise a lot of teams currently make.
Why this is arriving now
Guardrail spend has become a real line item, not an afterthought, as more products put an LLM directly in a user-facing critical path rather than behind a human review step. Lasso says the funding will go toward expanding engineering, deepening federal and regulated-industry work, and growing go-to-market in North America and Europe — the language of a company scaling into an established buying motion, not one still explaining why guardrails matter at all.
What this means for a small team's AI safety budget
If you currently sample-check AI outputs rather than inspecting every request, it is usually because inspecting everything on GPU-based tooling costs more than your margins support at your current traffic. A CPU-based option — from Lasso or a competitor that follows with something similar — is worth evaluating specifically against that constraint: does moving to full-coverage inspection become affordable at your actual request volume, not just in principle.
Ask what "CPU-based" actually means for latency before you switch anything. Compute type is one variable; added latency per request is the one that actually affects your user experience, and a vendor's cost claim does not automatically mean a speed claim holds too.
Do not assume full-coverage inspection is now free just because it is cheaper. Cheaper compute lowers the cost of a decision you still have to make deliberately — what to check for, what to block versus flag, and what a false positive costs you in a frustrated legitimate user versus what a missed real one costs you in an actual incident.
The broader pattern worth noticing
This is one company's product launch, not proof that AI safety tooling as a category has solved its cost problem — treat the specific performance and cost claims as unverified until independent benchmarks exist. But the direction is worth tracking regardless of which vendor wins: as more products put models on a genuinely critical, user-facing path, the pressure to inspect every request rather than a sample of them is only going to grow, and cost-per-check is exactly the lever that determines whether "inspect everything" is achievable for a team your size.
The bottom line
Lasso Security's $30 million round and its new CPU-based guardrail, LEAP, are a bet that AI safety inspection does not need GPU-scale compute to work at scale. Whether or not you ever use this specific product, the cost question it raises is a real one: if you are currently sampling instead of inspecting every AI request because of compute cost, it is worth checking whether that constraint has actually moved before assuming it has not.