Data security firm Fortanix Inc. has announced an expansion of its Fortanix Confidential AI platform, which now supports the Cloud Native Computing Foundation’s Confidential Containers project. This enhancement enables enterprises to run artificial intelligence workloads on Kubernetes within hardware-based trusted execution environments.
This announcement comes in response to risks associated with data exposure during inference, a phase in which model weights and input data must be decrypted in processor memory. At this stage, privileged administrators and cloud operators with access to the infrastructure could potentially access sensitive information.
Last week, Fortanix released survey results highlighting AI leaders’ perceptions of this risk. Conducted by Atomik Research, the survey polled 300 executives from large U.S. companies on behalf of Fortanix. An overwhelming 93% of respondents indicated that traditional encryption methods do not effectively protect data while AI is processing it. Moreover, 41% reported that security concerns had prevented them from deploying AI solutions altogether.
Fortanix’s solution builds upon the Foundations of Confidential Containers, which achieved incubating status with the CNCF in July. The open-source plugins being developed by the company ensure that a container’s legitimacy is verified before it is granted access to encryption keys. This verification process utilizes the Fortanix Confidential Computing Manager, which attests both CPU and GPU to cover inference jobs across both types of processors fully.
The Fortanix Data Security Manager only releases keys if the outcomes comply with the customer’s security policies, ensuring that they remain within a protected environment.
Anand Kashyap, Co-founder and Chief Executive of Fortanix, stated that transitioning AI workloads into containers “doesn’t change the threat,” as models and agents still interact with a company’s most sensitive data during operation. With the new plugins, deploying a container within a confidential computing environment can be achieved with a single click on compatible hardware and Kubernetes platforms.
Fortanix is particularly targeting regulated industries that require secure AI implementation in production environments. The key release operates on a cluster, ensuring that the loss of a single server does not impede operations. For customers with sovereignty requirements, the platform can be deployed on-premises or in an entirely air-gapped configuration. This capability also allows enterprises to host proprietary models from a frontier lab alongside their data without jeopardizing the lab’s intellectual property.
“Attestation is a protocol. Trust is an operation,” said Anuj Jaiswal, Fortanix’s Chief Product Officer. He emphasized that combining attestation and key management on a single platform alleviates the burden on enterprises to create their trust layer when adopting standard Confidential Containers.
Over a dozen companies, including Advanced Micro Devices Inc., Intel Corp., and CrowdStrike Holdings Inc., are supporting Fortanix in this initiative. John Fanelli, Vice President of Enterprise Software at Nvidia Corp., noted that Fortanix is leveraging open standards for attestation and secure key release “at production scale.” Nvidia and Fortanix previously launched a joint on-premises confidential AI platform in October of the previous year.
The Fortanix Confidential AI solution for Confidential Containers is now available. The company plans to showcase this technology on Nvidia GPUs with confidential computing capabilities at the Nvidia GTC Berlin event, scheduled for October 20-22.