Software Engineer Intern

Deepgram, a voice AI company, is powering automated conversational technology across commercial sectors including pharmacies, call centres, and air traffic control. The company's AI agents now handle high-stakes work, resolving over 90% of calls without human escalation. Deepgram's AI model, Flux, responds in 200 to 300 milliseconds and runs specialised models rather than relying on large language models. The technology can track conversations amid background noise, manage calls in multiple languages, and analyse caller emotions in real time. The company operates its own data centres and trains foundational models using the Dell AI Factory with Nvidia infrastructure. This includes Dell PowerEdge XE-Series servers with Nvidia-accelerated computing and Dell PowerScale storage. For regulated industries like healthcare and finance, Deepgram offers "air-gapped" systems where data remains entirely self-contained and never reaches the internet.

Deepgram partners with Fortanix and Nvidia for secure voice AI deployment in regulated industries. The collaboration brings confidential computing to real-time voice AI, keeping audio data and model weights encrypted even during active inference. Just now ago Voice AI has a trust problem in the industries that need it most. Hospitals, banks, and government agencies have spent years watching from the sidelines as consumer-facing AI tools race ahead, held back by a simple question: who else can hear what its AI hears? Deepgram is betting it has an answer. The voice AI company announced a partnership with Fortanix and Nvidia on June 1, 2026, to enable fully encrypted, on-premises voice AI deployments designed specifically for regulated industries. The core pitch: organizations can now run voice transcription, AI-powered customer support agents, and analytics tools while keeping both the audio data and the proprietary model weights encrypted during real-time inference. How confidential computing changes the game for voice AI. To understand why this matters, think about what normally happens when an AI model processes your voice. The audio gets decrypted so the model can analyze it, which creates a window of vulnerability. It's like a bank vault that has to open its doors every time someone needs to count the money inside. Confidential computing eliminates that window. The technology, built on Nvidia's Confidential Computing-enabled GPUs and powered by Fortanix's Confidential AI platform, allows Deepgram's models to process voice data while it remains encrypted. In English: the vault stays locked, but the money still gets counted. This applies to both sides of the equation. Sensitive audio, think patient conversations or financial advisory calls, stays protected. And Deepgram's proprietary model weights, the intellectual property baked into how its AI actually works, remain shielded from the organizations hosting them on-premises. Deepgram claims this is the first implementation of confidential computing in real-time voice AI applications. If that holds up, it positions the company at the front of a very specific but potentially enormous market: enterprises that want cutting-edge voice AI but can't or won't send sensitive audio to a third-party cloud. Building on an existing on-premises push. This announcement didn't come out of nowhere. Deepgram has been building toward secure deployment options for a while, offering on-premises and Virtual Private Cloud solutions as alternatives to its cloud API. The Fortanix and Nvidia partnership extends that strategy by adding a hardware-level encryption layer that goes beyond traditional network security. Just days earlier, on May 27, 2026, Deepgram announced a separate initiative around low-latency voice agents built on Nvidia's Nemotron technology. That release focused on speed. This one focuses on security. Together, they sketch out a product roadmap aimed at making Deepgram's voice AI fast enough and safe enough for the most demanding enterprise environments. The target use cases are straightforward: private on-premises voice agents that can handle customer interactions without data leaving the building, enterprise-grade transcription services for sensitive meetings and calls, and IT service desk tools that can automate support workflows while maintaining strict data governance. For healthcare organizations bound by HIPAA, financial institutions navigating a web of compliance requirements, and government agencies with strict data sovereignty mandates, these aren't nice-to-have features. They're prerequisites. For Fortanix, this collaboration extends their Confidential AI platform into voice, an area where the security stakes are arguably higher than in text-based AI. Voice data is inherently more personal, harder to anonymize, and often subject to stricter regulatory treatment than written communications. Disclosure: This article was edited by Editorial Team. For more information on how Cryptobriefing create and review content, see its Editorial Policy.

Deepgram has partnered with Fortanix to enable enterprises to run voice AI in on-premises environments with enhanced security protections. The solution combines Deepgram's voice AI models with Fortanix Confidential AI and NVIDIA Confidential Computing to protect sensitive data and proprietary model weights during active processing. The technology creates hardware-isolated environments where audio data and AI models remain encrypted throughout use, addressing security requirements in highly regulated industries like healthcare and finance. It enables organisations to deploy voice applications for customer interactions, transcription services and internal operations whilst maintaining data sovereignty and meeting HIPAA and GDPR requirements. The partnership aims to accelerate voice AI adoption in security-sensitive sectors by protecting data and models from theft or unauthorised access, even from privileged administrators.