Burlingame, CA, Nov. 10, 2026 (GLOBE NEWSWIRE) — DataKrypto today announced collaboration with Intel to integrate FHEnom for AI with next-generation AI infrastructure, based on Intel Xeon 6 processors and Intel GPU D-series, formerly code-named Crescent Island.
The collaboration addresses control of the trust boundary when enterprise and sovereign AI workloads run on infrastructure operated by a third party.
FHEnom for AI makes cryptographic protection portable with the workload. Model weights remain encrypted in storage, in GPU memory, during execution and as workloads move across supported infrastructure, while the accelerator holds no key material. This allows organizations to select compute based on performance, availability, power and cost while retaining cryptographic control of their AI assets.
Combining Intel Confidential Computing with Encrypted Execution
By combining Intel Trust Domain Extensions (Intel TDX) with FHEnom for AI, DataKrypto extends cryptographic protection to sensitive data and AI models across CPU and GPU execution.
- Intel TDX provides an attested trusted execution environment that helps isolate the workload on the Intel Xeon CPU from the host operating system, hypervisor, and infrastructure administrators.
- FHEnom for AI extends encryption to GPU processing, protecting sensitive data and AI model weights while computations are performed.
Together, Intel TDX and FHEnom help protect the workload across the execution environment while allowing the model owner to retain control of the keys and cryptographic boundary.
Customer Control of Model IP
Customers retain control over the key capable of recovering the model in plaintext. Depending on deployment requirements, that key can be sealed in trusted hardware, retained in the customer’s HSM, or destroyed after the model is encrypted.
In the latter configuration, the solution is designed to allow inference to continue on encrypted weights even though the key capable of reconstructing the original plaintext model no longer exists.
Encryption for Shared AI Infrastructure
FHEnom for AI enables multiple models, customers and workloads to share the same accelerator while maintaining independent cryptographic boundaries. Each model is protected with independently controlled keys, creating separate cryptographic domains for workloads running on the same hardware.
For cloud and infrastructure providers, this enables shared accelerator capacity while giving each customer independent cryptographic control over its model.
Protecting Prompts Across Client Architectures
FHEnom for AI supports two client architectures, allowing organizations to determine where encryption begins.
- Thick Client: FHEnom encryption begins on the customer’s device or within the customer’s controlled environment. Prompts, documents, RAG content and other sensitive information are encrypted before leaving that environment, so the infrastructure provider does not receive the original information in plaintext.
- Thin Client: Standard TLS communication terminates inside an attested Intel TDX environment, where plaintext is transformed and encrypted under FHEnom before being sent to the GPU for execution.
In both configurations, model weights and protected intermediate information remain encrypted during GPU execution, keeping model protection independent of the client architecture and underlying accelerator.
“Sovereign AI should mean more than owning the hardware or knowing where the data center sits. It should mean holding the keys, or deciding that no key exists at all,” said Luigi Caramico, Founder and CTO of DataKrypto. “When confidentiality travels with the workload, organizations can choose infrastructure for performance and economics while keeping control of the models and data that create their competitive advantage.”
“As AI models become some of the most valuable digital assets organizations own, protecting them throughout execution is becoming increasingly important,” said Arijit Bandyopadhyay, CTO of Enterprise Analytics & AI, Data Center Group, Intel Corporation. “DataKrypto has developed a unique approach to AI security that keeps models and sensitive information cryptographically protected during GPU execution. By combining FHEnom for AI with Intel Xeon, Intel TDX and the Intel GPU-D series technology, we are working together toward a new level of security for enterprise and sovereign AI infrastructure.”
About DataKrypto
DataKrypto advances confidential AI through encrypted execution. Its FHEnom for AI™ technology protects sensitive data, AI models, prompts, and intermediate computations during training, fine-tuning, and inference, including GPU processing. By extending encryption to AI workloads while in use, DataKrypto enables organizations to protect proprietary intelligence across the AI lifecycle.
For more information, visit datakrypto.ai/partners/intel
©Intel, the Intel logo, and other Intel marks are trademarks of Intel Corporation or its subsidiaries
Press Inquiries
Michelle Stone
michelle.stone@datakrypto.com
https://datakrypto.ai
Media gallery
