AI SecurityData Protection
Cloaked AI
Encrypts vector embeddings while keeping nearest-neighbor search and clustering usable.
Cloaked AI Overview
What it does
Cloaked AI is a vector encryption software development kit (SDK) that protects the embeddings behind retrieval-augmented generation (RAG), semantic search, biometric matching, and recommendation systems, where inversion attacks can reconstruct source text, images, and personal data from stored vectors. It applies approximate distance-comparison-preserving encryption, a property-preserving scheme that keeps ciphertext vectors usable for nearest-neighbor search and clustering, so embeddings remain encrypted inside the vector database instead of being decrypted for each query.
How it works
Cloaked AI encrypts each vector element by multiplying it by a secret scaling factor, then adding pseudorandom perturbation bounded by a configurable approximation factor that trades search accuracy against how hard the original point is to guess, and finally shuffling the elements deterministically. Keys and scaling factors are scoped per data segment, typically a single tenant, so cluster analysis across segments reveals nothing. Metadata fields receive deterministic encryption when they are used to look up vectors and standard encryption otherwise. Applications call encrypt before storing and before querying, using the open-source ironcore-alloy library in Rust, Python, Java, and Kotlin.
Credentials and traction
IronCore Labs holds annual SOC 2 Type 2 certifications for Security, and Cloaked AI was named in Gartner Cool Vendors in Data Security 2025, cited for encrypting the vector embeddings that carry sensitive data through AI pipelines. Broadcom, HubSpot, Zendesk, and Norwegian Cruise Line Holdings are IronCore Labs customers. Cloaked AI targets engineering teams in regulated sectors, including healthcare, financial services, and biometrics, that need private data inside generative AI systems without exposing it to the vector database operator.
Key Capabilities
mapped to solution categoriesIsolates embeddings by tenant and user so an unauthorized query returns nothing, preventing cross-tenant or cross-user leakage of retrieved content.
Compliance
certificationsIntegrations
compatible toolsImplementation & support
Info last updated on July 26, 2026
Vendors
Is this your product?
Claim your profile to connect with the teams looking for your solutions.