OC3 registrations are now open! Join the premier event for confidential computing online or in Berlin on March 27.
Blog
Lara Montoya Laske
Organizations and public institutions can greatly benefit from adopting AI to streamline processes, enhance customer experiences, and gain valuable insights through advanced data analytics. However, for many applications, concerns over data protection and privacy remain a major barrier. Confidential AI provides cutting-edge solutions to address these challenges.
Confidential AI leverages confidential computing technology to secure sensitive data at unprecedented levels throughout the entire interaction with AI models. This is achieved by ensuring that data remains encrypted at all times—not only at rest or in transit but also while being actively processed. Although confidential computing has been used for some time to protect large-scale workloads on CPU-based infrastructures, it couldn’t previously be extended to AI accelerators like GPUs. The limitations stemmed from the underlying hardware architectures of AI accelerators, which prevented them from establishing a protected channel with a secure enclave or confidential virtual machine (CVM). This barrier has been overcome with the introduction of novel accelerators, such as Nvidia H100 GPUs. These GPUs, for the first time, offer hardware-enforced trust mechanisms, making Confidential AI a reality.
Rooted in the strong security properties of confidential computing, Confidential AI offers the perfect solution to leverage AI securely without exposing sensitive information.
A state-of-the-art implementation of Confidential AI is Continuum. Continuum is a generative AI service similar to ChatGPT but uses confidential computing to keep all user data private. It offers an intuitive API that can act as a drop-in replacement for OpenAI, enabling automated processing of sensitive data. The browser version includes a familiar chatbot interface for easy interaction.
For more on Confidential AI and Continuum, read our whitepaper.
As industries increasingly adopt AI for decision-making, automation, and other applications, Confidential AI can serve as a catalyst for business processes and use cases that require strong data security. This is especially important in sectors like finance, healthcare, and the public sector, where data privacy and regulatory compliance are vital. Below are some exemplary use cases that can benefit from Confidential AI:
Finetuning AI models for specific company needs
Many companies want to finetune AI models for their specific know-how to make them more efficient for their use cases. Finetuning AI models requires access to large volumes of company data, often including sensitive customer information or intellectual property (IP). Confidential AI allows businesses to finetune models entirely on encrypted data, ensuring that the data remains secure during the entire training process.
Building on the previous use case, AI can be used for secure and efficient data analysis. One example of an AI-based data intelligence tool is JP Morgan's "COIN" system. Traditionally, their legal teams would spend 360,000 hours annually reviewing documents. COIN, once fine-tuned for the task, handles the same volume in mere seconds. However, documents containing personal customer information must be strictly protected from any third-party access. With Confidential AI, organizations could extract insights from large datasets without compromising privacy.
In addition to finance, healthcare is, of course, a highly sensitive and tightly regulated sector. Patient data, combined with large medical databases, can be processed and analyzed by AI models to provide tailored insights into specific conditions or illnesses. This enables medical institutes and organizations to leverage AI for valuable insights without violating the privacy of patients.
AI-powered chatbots are a great tool for customer support and internal assistance. With Confidential AI, all interactions between customers and chatbots, or between employees and internal bots, are fully encrypted. This prevents data leaks and potential compliance breaches.
Multi-party collaboration
Confidential AI enables secure collaboration across organizations without the need to share sensitive data. In industries like banking or research, multiple parties can work together on AI-driven projects without exposing proprietary or confidential information. This allows for the development of shared insights while maintaining control over sensitive data.
For example, financial institutions can use Confidential AI to improve fraud detection and risk management. By analyzing transaction data between banks with confidential computing, AI models can identify fraudulent activities while still maintaining the confidentiality of customer information. Data remains inaccessible to unauthorized users, including third-party service providers, while AI models can identify suspicious patterns.
Confidential AI is a key tool in helping companies meet stringent data protection regulations like GDPR, DORA, and HIPAA. It enables secure processing of sensitive data—whether for document validation, auditing, or compliance checks—without exposing it to third parties. By encrypting data at all times, Confidential AI supports businesses to meet compliance standards and avoid costly penalties for data breaches.
In sectors where AI models represent valuable intellectual property, Confidential AI ensures that proprietary algorithms and models are kept 100% secure. Encrypting the model weights themselves and keeping them hidden from the service and cloud providers prevents misuse of the IP, even if the models are deployed in public clouds or any untrusted environment.
With our Continuum API, you can easily build secure and compliant AI services to support your company’s needs. Interested in having a chat with our experts? Contact us. For technical info on Continuum, you can visit the documentation. For more on Confidential AI, we also offer a comprehensive whitepaper.
Author: Lara Montoya Laske
