What is Cloaked AI?

IronCore Labs offers Cloaked AI, a solution designed to address the security risks associated with using sensitive data in generative AI applications, particularly those employing Retrieval-Augmented Generation (RAG) and vector databases. When private data is converted into vector embeddings for AI processing, it creates 'shadow copies' that, while not human-readable, can potentially be reverted to approximate the original sensitive information. Cloaked AI mitigates this risk by encrypting these vector embeddings directly.

This encryption process ensures that the vector data remains protected even if the storage system is compromised. Crucially, Cloaked AI allows these encrypted embeddings to be securely searched and utilized by AI models without needing to decrypt the entire dataset, maintaining functionality while enhancing data privacy and security. It integrates with various vector databases, providing a robust layer of protection against data breaches and unauthorized access related to AI vector data.

Features

Vector Embedding Encryption: Encrypts sensitive AI data represented as vector embeddings in databases.
Queryable Encryption: Allows searching and querying of encrypted vector embeddings without decryption.
Functionality Preservation: Ensures encrypted data remains usable for AI model processing.
Sensitive Data Protection: Safeguards against breaches and unauthorized access to AI-generated shadow data.
RAG Security Enhancement: Mitigates risks associated with using private data in RAG architectures.
Metadata Security: Secures associated metadata alongside vector embeddings.

Use Cases

Securing generative AI applications using private or sensitive data.
Protecting customer data processed by AI models.
Enhancing data privacy in RAG-based AI systems.
Implementing secure search over sensitive vector embeddings.
Meeting compliance requirements for AI data handling.
Securing AI applications in regulated industries.

FAQs

What are the risks of using vector embeddings in AI?

Vector embeddings are shadow copies of original data and can potentially be inverted back to approximate sensitive source material, posing a security risk if not protected.
How does Cloaked AI protect vector embeddings?

It encrypts the vector embeddings, making them unreadable without the correct key, while still allowing them to be searched and used by AI applications.
Can you search encrypted vector embeddings with Cloaked AI?

Yes, Cloaked AI uses techniques like queryable encryption (partially homomorphic encryption) to allow searching over encrypted vector data.
Is Cloaked AI suitable for RAG architectures?

Yes, it is designed to mitigate the data security risks inherent in RAG architectures that leverage private or sensitive data.
Does IronCore Labs address post-quantum cryptography concerns?

Yes, IronCore's platform is designed to be crypto-agile, allowing changes in cryptographic algorithms over time to adapt to threats like quantum computing.