Capeprivacy

As organizations increasingly adopt Artificial Intelligence (AI), particularly Large Language Models (LLMs), for critical operations, ensuring data security and privacy becomes paramount. Capeprivacy addresses these challenges through sophisticated encryption and secure processing technologies while providing flexibility in deployment and integration options. This technical overview explores Cape's encryption framework, secure data processing mechanisms, and its implementation for AI-driven applications, demonstrating how the company enables safe AI adoption across enterprise environments.

Overview of Cape's Encryption Tools

Cape provides two core encryption tools: cape encrypt and cape run. These tools enable secure data processing within Cape's secure enclave while abstracting users from complex security management tasks.

Encryption Process

When encrypting data using cape encrypt, the process begins with the generation of a 256-bit data encryption key (DEK). The input data is then encrypted with this DEK using AES-256. The DEK itself is encrypted with the user's Cape Key using RSA-4096. The combined AES and RSA encrypted data forms an encryption envelope, which is base64-encoded and prefixed with cape: before being presented to the user. This entire process occurs offline on the local machine, ensuring data remains entirely within the user's control.

Key Management

Cape handles key management through its integration with AWS Key Management Service (KMS). When cape encrypt is first run, it generates a unique key pair within the KMS, consisting of a public key used for encryption and a private key responsible for decryption. The private key remains immutable and never leaves the KMS, maintaining an additional layer of security. The public key is transmitted to the user's system as part of the attestation process, where it is verified and considered safe for data encryption.

Data Processing Security

After data storage, encryption remains a critical aspect of data processing within Cape's secure enclave. When cape run is executed, it initiates the computational process in Cape's secure enclave, a Hardware-Enabled Trusted Execution Environment (TEE). The enclave undergoes a rigorous attestation process to verify its identity, utilizing platform configuration registers (PCRs) to ensure the execution environment matches expectations. This process, combined with stringent certificate validation and PCR verification, ensures that only authorized and properly configured enclaves can decrypt and process data.

Network and Offline Security

The encryption process is designed to maintain security even in offline environments. Initial network requests for the Cape Key occur only once during setup, after which subsequent calls to cape encrypt do not require network access. This capability allows users to encrypt local data behind firewalls before storing it in cloud environments. The entire encryption and processing workflow operates within the secure enclave, protecting data confidentiality during both storage and transit.

Cape Encrypt Functionality

When developing applications that process sensitive data, developers often face complex challenges related to encryption key management and secure data processing. Cape's cape encrypt tool simplifies this process by securely managing encryption keys and handling the complexities of secure data processing.

The encryption process begins with the generation of a unique key pair within Cape's AWS Key Management Service (KMS). This key pair consists of a public key used for encryption (the Cape Key) and a private key responsible for decryption. Notably, the private key remains immutable and never leaves the KMS environment, providing an additional layer of security.

When encrypting data, cape encrypt generates a 256-bit data encryption key (DEK) and uses AES-256 to encrypt the input data. The DEK itself is encrypted using the RSA-4096 algorithm with the user's Cape Key. The combined AES and RSA encrypted data forms an encryption envelope, which is base64-encoded and prefixed with cape: before being presented to the user. This entire process occurs offline on the local machine, ensuring that data remains entirely under the user's control throughout the encryption process.

To facilitate secure data processing within Cape's secure enclave, the encryption process employs multiple layers of security verification. The enclave undergoes a rigorous attestation process to prove its identity, utilizing platform configuration registers (PCRs) to ensure the execution environment matches expected configurations. This process, combined with detailed certificate validation and PCR verification, ensures that only properly configured and authorized enclaves can decrypt and process data.

The secure key management system further enhances security through its implementation details. The private key never leaves the KMS environment, and the public key is securely transmitted to the user's system as part of an attestation document signed by the enclave. The client verifies this document, checking both the signature and a unique nonce value, before considering the key safe for data encryption. This process maintains strict control over key distribution and usage, ensuring that encryption keys remain secure and inaccessible to unauthorized parties.

Data Processing Security Mechanisms

Cape's secure data processing environment operates within a Hardware-Enabled Trusted Execution Environment (TEE), specifically utilizing Intel SGX technology for their secure enclaves. This TEE serves as a secure foundation for all data processing operations, isolating execution from the broader system while maintaining close integration with system resources.

Attestation forms a critical security mechanism in the data processing pipeline. Before any data processing can occur, the secure enclave undergoes a rigorous identity verification process. This attestation process employs a combination of cryptographic hashes and Platform Configuration Registers (PCRs) to confirm the enclave's identity. The enclave's unique configuration is compared against the expected measurements, with any discrepancies immediately preventing access to decrypted data.

To ensure the integrity of their key management infrastructure, Cape employs specific security measures within their Key Management Service (KMS) implementation. The service scope is carefully tailored to each Cape account, employing condition keys that govern permitted operations. Under Cape's configuration, the KMS enforces restrictions against PCRs 3 and 8, limiting what cryptographic operations can be performed.

The data processing flow incorporates multiple layers of security validation. When a service requests attestation to a specific enclave, this process begins with a TLS connection, where the enclave's certificates undergo rigorous validation. These certificates contain crucial information that must match the expected configuration for access to be granted. Once authenticated, the service uses the enclave's public key to encrypt data, which is then processed within the secure enclave.

Data processing within the secure enclave strictly adheres to a controlled workflow. The encrypted data arrives in the enclave via HPKE (HyperText Post-Quantum Key Encapsulation), protected by TLS encryption. Inside the enclave, the process unfolds as follows: the enclave's private key decrypts the HPKE-protected data, which is then securely processed using AWS KMS for additional storage encryption. This double layer of encryption ensures that data remains inaccessible to unauthorized parties throughout its lifecycle within the secure processing environment.

Integration and Deployment Options

Capeprivacy offers flexible deployment options to meet various enterprise needs. The platform supports both on-premise and Virtual Private Cloud (VPC) deployments, allowing customers to choose the most secure and efficient configuration for their environment.

The company provides secure access to both public and private Large Language Models (LLMs), including support for OpenAI's GPT-4 and Anthropic's Claude-3. This flexible model selection enables businesses to leverage the most appropriate AI technology for their specific use cases.

Cape's integration capabilities support custom models and workflows, allowing developers to build tailored applications that meet their organization's unique requirements. The platform's modular architecture enables seamless integration with existing systems and tools, including document management platforms like Google Docs, GitHub, Slack, and Zoom.

The company's technology stack includes multiple AI models, with processing capabilities spanning over 350,000 redactions across their platform. This significant data processing volume demonstrates the company's ability to handle complex AI workloads while maintaining high levels of security and privacy.

CapeChat Plus subscribers gain direct access to GPT-4 and enhanced features for prompt customization. The platform implements usage-based pricing through tokenized billing, where 100 tokens represent approximately 75 English words. This structure enables users to manage costs effectively based on their specific needs.

The platform's technical architecture emphasizes security throughout the data processing workflow. When uploading documents, Cape employs advanced encryption techniques including Intel® Total Memory Encryption and AWS Nitro Enclaves. All network communication uses TLS encryption to protect data in transit.

The company's approach to data redaction and privacy protection incorporates state-of-the-art techniques for entity removal across multiple language variants. Users can access a comprehensive suite of privacy tools for both de-identification and re-identification processes, supporting redaction for 50 specific entity types across 47 languages.

Cape Chat and Retrieval Augmented Generation

CapeChat Plus subscribers have direct access to GPT-4, which powers the platform's natural language processing capabilities. The service uses advanced privacy techniques to protect user data, including state-of-the-art entity removal across multiple language variants. Supported file types include PDF, Word, Excel, PowerPoint, CSV, Text, and Markdown, allowing users to upload multiple documents for focused conversations.

All data remains under user control, with CapePrivacy and AWS sub-processors having no access to plaintext information. Data security employs multiple layers of protection, including Intel Total Memory Encryption, AWS Nitro Enclaves, AES 256, RSA 4096, and AWS KMS. Sensitive data undergoes thorough cleansing before processing through OpenAI's ChatGPT API, while all network communication is encrypted using TLS.

The platform supports integration with various third-party tools through its API, including Google Docs, GitHub, Slack, and Zoom. The service demonstrates effectiveness across multiple industries, from financial data interpretation to automated policy document answering and structured data conversion. The free version enables initial exploration, while the Plus plan provides advanced features including direct GPT-4 access, full prompt control, and direct API integration. Usage follows a per-token model, with 100 tokens representing approximately 75 English words, providing both cost-effectiveness and flexibility.