LLM Security & Privacy

BlueNexus employs confidential computing technologies — including hardware-based memory encryption and runtime attestation — to provide verifiable isolation for sensitive workloads.

When using a BlueNexus-hosted model:

• Your prompts are encrypted during transit and during execution in memory.

• The underlying hardware (CPU/GPU) enforces encryption at the chip level.

• No logs or metadata can be accessed by BlueNexus staff, system administrators, or cloud providers.

This ensures a zero-trust execution environment, with end-to-end data protection guarantees.

When using a Proprietary-hosted model, BlueNexus is unable to provide any guarantees around privacy or security.

Open-Source Models (Confidential Execution)

In addition to routing to proprietary models, BlueNexus also hosts and operates open-source models under a secure, privacy-first architecture.

These models run in Confidential Mode, within Trusted Execution Environments (TEEs) that ensure:

• In-memory encryption of both CPU and GPU computations.

• No human or system-level visibility into your prompts or responses.

• No data persistence — prompts are not logged or used for training.

• End-to-end confidentiality for every API call.

This architecture provides strong privacy guarantees, making these models suitable for:

• Healthcare

• Finance

• Legal and compliance-driven applications

• Government and defense-grade environments

In essence, your data remains yours. BlueNexus (or any third party) cannot access or reuse it.

Billing and performance tracking

Confidential LLM requests trigger recording of the following information; Token usage (prompt, completion, total), Model used, Confidential status of model, Request performance metrics (timeouts, response time).

Proprietary Model Access

BlueNexus provides transparent access to proprietary models such as GPT (OpenAI), Claude (Anthropic), and other vendor APIs.

When you select one of these models:

• Your requests are securely routed through BlueNexus to the underlying provider.

• Prompts and responses are processed in line with the provider’s own terms of use and privacy policies.

• These third-party providers may retain or analyze prompts for model improvement, safety monitoring, or other operational purposes.

Because of this, you do not have full control over the storage, handling, or retention of your users’ prompts and responses. If your application operates in a regulated industry or requires strict data privacy guarantees, you should consider using BlueNexus’s confidential hosted models (see below).