Fortanix, a worldwide top company in data and AI security and one of the first to promote Confidential Computing, has launched a new Confidential AI solution based on NVIDIA Confidential Computing. This new product is aimed at helping companies securely install and operate AI models, either their own or those of third parties, in their own environment without releasing their intellectual property or other sensitive data.
The product features one of the most serious obstacles on the way of AI implementation for enterprises namely the dual need for protection of model ownership along with data privacy. Thanks to the use of Confidential Computing, it is now possible to perform high-level AI tasks using sensitive information without this data being made available to model providers and at the same time to ensure that proprietary models are kept safe from being accessed or misused by unauthorized persons.
Enabling Secure AI Inference at Scale
Thus, the Fortanix Confidential AI, when integrated with NVIDIA Confidential Computing, enables enterprises to securely process AI inference workloads. In the process, the model weights remain encrypted and inaccessible, even to the infrastructure. Meanwhile, the sensitive data remains processed securely.
This enables the enterprises to leverage the best models of AI, while at the same time maintaining full control over the data. In the process, the developers of the models are also able to distribute the models, thus generating revenue without the risk of replication or intellectual property theft.
Also Read: Experis and SoundHound AI Announce Partnership on Enterprise AI Adoption
Cryptographic Assurance for Data and Model Protection
The platform introduces verifiable, cryptographic guarantees that replace traditional trust models based on contracts alone. Key security capabilities include:
Controlled release of cryptographic keys only to verified and trusted runtime environments
Protection of model weights from tampering or unauthorized access
Encryption of customer prompts and outputs during processing
Continuous validation of the deployment environments to detect unauthorized changes
Through the use of Confidential Computing and secure key management and centralized policy enforcement, the solution provides end-to-end security for the AI workloads during runtime.
Unlocking Trusted Collaboration Between Enterprises and Model Providers
The new solution enables a secure ecosystem where enterprises and AI model providers can collaborate with confidence. Organizations can deploy third-party models on highly sensitive datasets—such as regulated or proprietary information—without violating privacy or compliance requirements. Meanwhile, model developers can expand their reach into enterprise environments without exposing their intellectual property.
“AI security can break during inference if you don’t protect data and models in use,” said Anuj Jaiswal, Chief Product and Strategy Officer, Fortanix. “This deployment on NVIDIA Confidential Computing-backed GPUs, verified by Fortanix Confidential Computing Manager and backed by secure key release from Fortanix Data Security Manager, demonstrates that you no longer have to choose between performance and protection. Confidential AI enables both.”
Enabling Secure and Scalable Adoption of AI
By resolving critical issues related to data sovereignty, privacy, and intellectual property rights, Fortanix’s Confidential AI solution is opening doors for wider enterprise adoption of AI. Now, enterprises are able to confidently utilize third-party models for driving innovation and business results while maintaining tight security controls in place.
For AI model providers, Fortanix’s solution also offers a secure way to scale and commercialize their offerings in on-premises and enterprise markets.