Analysis
Lasso Security raised $30 million in a round led by ClearSky, with Entree Capital -- which led the company's seed round -- increasing its position, alongside iAngels, Singtel Innov8, Mindset, and Swish Data, SiliconANGLE reported. The round takes total investment in the company to more than $37 million.
What Lasso Actually Sells
Founded in 2023 by Schulman and Dror, Lasso builds guardrail systems that screen the inputs and outputs of AI models for security and safety violations -- prompt injection, data exfiltration, jailbreak attempts, and the kind of policy violations enterprises need blocked before an LLM response reaches a customer or takes an action. The new capital funds LEAP, a transformer-free class of guardrail model the company says delivers top-tier detection accuracy running entirely on ordinary CPUs, at under five milliseconds per decision and, Lasso claims, thousands of times the throughput of existing GPU-based guardrails.
“That CPU-only claim is the commercial pitch, not a technical curiosity.”
That CPU-only claim is the commercial pitch, not a technical curiosity. Every enterprise running guardrails on today's transformer-based detection models is paying GPU-scarce, GPU-priced compute just to screen traffic before it ever reaches the model doing the actual work -- a tax that scales linearly with usage in a year when GPU capacity is the industry's tightest constraint. A guardrail that runs on commodity CPUs instead removes that tax, which is why the pitch resonates specifically with the federal and regulated-industry accounts Lasso says this round is earmarked to pursue.
Timing, Not Coincidence
Lasso's raise lands the same week OpenAI rated its Astra model a first-ever "Critical" cybersecurity risk and CrowdStrike launched SafeMind, its own offense/defense AI security system. AI-security guardrails and red-teaming tooling are becoming one of the most active sub-categories in the funding market precisely because frontier labs are now publicly grading their own models as capable of autonomous exploit-finding -- every enterprise deploying those models needs a screening layer it can trust, and few want to build one in-house.
The unresolved question is defensibility. Guardrail detection is a moving target -- jailbreak techniques evolve constantly, and a CPU-only architecture has to prove it keeps pace with GPU-based competitors on accuracy, not just cost, or the latency advantage stops mattering. Lasso's next test is landing named regulated-industry customers willing to say so publicly, the kind of reference accounts that convert a good demo into a durable enterprise security category.
