Lasso Security — LEAP CPU-based guardrails and $30M round (September 2026)
Tag: S-2026-09-02-lasso-leap-cpu-guardrails Type: article (primary — vendor press release via GlobeNewswire, fetched in full) Author(s): Lasso Security Date of source: 2026-09-02 Date ingested: 2026-09-08 Authority weight: medium — primary on the product announcement and funding facts; every performance, accuracy and growth claim is vendor-asserted and not independently corroborated in this scan Raw file: /_raw_sources/S-2026-09-02-lasso-leap-cpu-guardrails.md
What it claims
Lasso Security (LLM/GenAI security vendor on Paul’s guardrails watchlist) announced LEAP, described as “a new transformer-free class of AI guardrail” that it claims 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”. The deployment argument is locality: “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.”
Alongside, Lasso announced a $30 million funding round led by ClearSky with participation from Entrée Capital, to “expand engineering, deepen its federal and regulated-industry work, and grow go-to-market in North America and Europe”. It claims revenue growth of over 500% in 12 months and names representative customers: US Department of Homeland Security, Kaufman Rossin (accounting), eToro (online broker/trading platform), Fiverr; it also names BMW and Leonardo Defense, and Protiviti as a channel partner. A red-team anecdote states that at “a global healthcare provider” Lasso’s red team “extracted all of the patients records” — offered as evidence of the risk LEAP addresses.
Notable quotes
- “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” (release, para 1)
- “LEAP deploys anywhere the customer’s data has to stay, be it the customer’s own cloud, regulated environments or air-gapped networks.” (release body)
- “representative customers including US Department of Homeland Security, Kaufman Rossin, eToro, Fiverr and more” (release body)
- “a $30 million funding round led by ClearSky, with participation from Entrée Capital” (release body)
What’s speculative vs. asserted
Asserted (verifiable in principle): the LEAP launch, the $30M round and its investors, the named customers and partner.
Vendor-asserted, unverified: “top-tier detection accuracy”, <5 ms latency, “thousands of times the throughput”, 500% revenue growth, the healthcare red-team anecdote (unnamed target). No benchmark, dataset or third-party test is cited. “Transformer-free” is a design claim without technical detail in the release.
Notably absent: no regulatory standard named (EU AI Act, ISO 42001, NIST AI RMF, DORA); no statement on evidence/logging artefacts produced by LEAP decisions; no EU entity named despite the Europe go-to-market intent.
Topics this feeds
- AI Governance Platforms — GenAI guardrail cohort; data-locality (CPU-only, air-gapped) as a differentiation axis; eToro as a rare named FS-adjacent reference.
Open questions raised
- Can LEAP’s accuracy/latency claims be independently benchmarked (e.g. against transformer-based guardrails on public prompt-injection sets)?
- Is eToro a production deployment and in which regulated entity/jurisdiction (eToro operates under FCA, CySEC and other licences) — not stated.
- Does a CPU-only, in-boundary guardrail change the DORA/GDPR analysis for GenAI in FS (data stays in-house) versus cloud-hosted guardrails — this vault’s inference, not a source claim.