Artificial intelligence / news brief
Cohere opens confidential AI inference on Azure H100 GPUs in public beta
- Event
- Published
Cohere has added a public beta of confidential computing to Model Vault, its managed single-tenant inference service. The system uses Azure confidential virtual machines, NVIDIA H100 confidential computing, remote attestation and encrypted payloads so customers can verify the environment before releasing sensitive requests.
Why it matters
Hardware-backed isolation can reduce who must be trusted with plaintext during hosted inference, though the beta has no published external audit or performance overhead data.
Sources
- Introducing Confidential Computing on Cohere Model Vault (external site)Cohere · primary source ·
- Cohere's Model Vault now encrypts AI inference so even Cohere cannot see enterprise customers' data (external site)VentureBeat · coverage ·