AI is fundamentally changing what organizations need to protect. For years, enterprise security teams have focused on protecting data when it is stored and when it moves across networks. But AI creates a different problem: sensitive data is now being processed by models, agents, retrieval systems and inference pipelines that often run on shared infrastructure.
Confidential AI is the application of confidential computing to AI workloads. It uses hardware-backed trusted execution environments to help protect data and workloads during processing. In practice, this means running AI applications in environments where the host operating system, hypervisor, cloud administrator, or infrastructure operator have reduced access to the workload and its memory.
Confidential AI can protect several assets that are central to modern AI systems:
- User input: Prompts may include personal data, business context, proprietary source code, customer information, or regulated records.
- Retrieval context: RAG systems often connect models to internal documents, databases, and knowledge bases. The data retrieved for a prompt may be more sensitive than the prompt itself.
- Model weights: For model providers, weights are intellectual property. For enterprises, fine-tuned models may encode business-specific expertise or regulated data patterns.
- Inference output: Outputs may contain summaries of confidential data, generated decisions, extracted personal information, or sensitive recommendations.
- The agent runtime: Agentic AI makes the problem harder because agents may hold memory, credentials, API tokens, tool outputs, and long-running task states.
As enterprises move from AI experiments to production systems, that foundation will matter more. Confidential AI gives organizations another option, and is central to making AI infrastructure trustworthy enough for the workloads organizations actually want to run.
Learn more
- Ubuntu for confidential compute
- Ubuntu for confidential AI
- Get your Ubuntu Pro subscription
- Talk to an expert about your open source needs
Download the whitepaper to learn more!