Field notes from
confidential AI.
Technical deep dives, product updates, and engineering notes from the Ultraviolet team on building open-source confidential computing and sovereign AI.
Beyond the Identity Crisis: Level 2 Binding in Cocos aTLS
We've overhauled our attested TLS implementation to achieve Level 2 session binding, formally mitigating the relay attack vulnerabilities identified at IETF SEAT.
Cube AI Public Beta: The Future of Confidential LLM Inference is Here
We are incredibly excited to announce the public beta release of Cube AI. Unlock the power of open-source, hardware-secured generative AI. Build and deploy agentic workflows with absolute data privacy using Trusted Execution Environments.
TLS and TEEs: Cocos AI Attested TLS — Design, Limitations, and the Path Forward
A rigorous look at Cocos AI's attested TLS implementation, the relay attacks identified by formal analysis at IETF SEAT, and what we plan to do about them.
AMD SEV-SNP vs Intel TDX: Understanding TEE Support in Cube AI
A technical comparison of AMD SEV-SNP and Intel TDX confidential computing technologies, and how Cube AI abstracts their complexities to provide a unified, secure platform for AI workloads.
Securing Trust: RATS and Attested TLS (aTLS)
COCOS AI leverages modern attestation standards to prove the integrity and operational state of the platform.
The Privacy Paradox: Why AI Teams Should Care About Confidential Computing
AI teams face a growing privacy paradox: they must process highly sensitive data at cloud scale while traditional architectures leave data in use exposed. This post explains how confidential computing and TEEs close that gap for enterprise AI workloads.
Prism AI Public Beta: Empowering Secure AI Collaboration
We are thrilled to announce the public beta release of Prism AI, our confidential computing platform for secure AI collaboration. Build and deploy privacy-preserving AI with hardware-verified security.
vLLM vs Ollama in Cube AI: Choosing the Right LLM Backend for Your Use Case
Compare vLLM and Ollama as pluggable LLM backends in Cube AI, and learn how to choose the right engine for your latency, throughput, cost, and deployment requirements.
Unleashing Confidential AI: Cocos v0.8.0 and Prism v0.6.0 Released
Ultraviolet releases Cocos AI v0.8.0 and Prism v0.6.0, enhancing the confidential computing ecosystem. These updates bring improved attestation, stronger security, and a better enterprise experience.
Confidential Computing Meets AI: How Cube AI Protects Your LLM Prompts
Cube AI leverages confidential computing and TEEs to protect your sensitive AI workloads. Ensure your prompts remain private with hardware-verified security.