According to the AWS Machine Learning Blog, the company Octave built a secure multi-tenant AI foundation on Amazon Bedrock. It supports the Complaint Triage Assistant—a helper for classifying complaints, which performs searches across 2 billion regulated documents without crossing boundaries between tenants.
The description also notes that Octave used Cohere Embed v4 to create vector embeddings, and Amazon Bedrock AgentCore for isolated execution of assistant tasks.
The practical significance of the solution is the combination of large-scale search over sensitive materials with separation of data for different clients. The source does not report the system's accuracy, launch timelines, or actual results.
editorial commentary
Why it matters
If the claimed isolation is confirmed in practice, such an approach could simplify deploying AI in tasks with regulated data. The next observable signals will be independent information about accuracy, operation and checks of tenant separation. A significant uncertainty is that the conclusion is based on one synopsis of the AWS Machine Learning Blog without independent confirmation.