RefortifAI

RefortifAI, a Y Combinator Spring 2026 company based in San Francisco, builds a runtime-level protection layer for AI model weights. The core idea is to let a model owner ship a proprietary model to a partner, licensee, or deployment environment while keeping the underlying weights unusable outside RefortifAI’s hardened runtime, even if the model files themselves are extracted or copied.

The product is designed as a drop-in replacement for common inference engines such as vLLM and SGLang, so customers can deploy it without re-architecting their serving stack. A companion CLI reportedly handles the weight-obfuscation step before deployment. The target market is organizations that license specialized or proprietary models to third parties, including defense contractors and enterprises sharing fine-tuned models with partners, where the risk of model extraction or unauthorized reuse is a direct business and IP concern.

RefortifAI is very early stage: public information is limited to Y Combinator’s own listing and a handful of aggregator pages, with a reported team of about three people. There is no independently verified customer list, benchmark data, or disclosed funding beyond YC’s standard investment, so claims about the robustness of its weight-obfuscation approach have not yet been tested outside the company’s own description.