Lasso Security launches LEAP: CPU-based AI security guardrails with accuracy on par with GPU solutions

Lasso Security released LEAP, a set of CPU-based AI security guardrails, claiming document processing throughput roughly three orders of magnitude higher than commercial guardrail APIs and closed-source LLM judges, while maintaining GPU-level accuracy. LEAP targets the cost and latency issues of placing an LLM judge before every prompt, response, retrieved document, and tool call: a single GPU instance (e.g., AWS ml.g7.2xlarge, $3.15/hour) running continuously costs about $2,300 per month. Lasso argues that low-ambiguity inputs do not require open-ended reasoning and that reasoning models should be reserved for genuinely ambiguous security decisions.