World's Fair 2025
GPU-less, Trust-less, Limit-less: Reimagining the Confidential AI Cloud - Mike Bursell
Overview
This talk introduces confidential AI as a solution to trust issues in AI development and deployment. It explains how confidential computing, utilizing Trusted Execution Environments (TEs), protects data and models during processing, even from system administrators or hardware providers. This technology enables secure collaboration, monetization, and the use of sensitive data for AI tasks across various industries.
Who should watch
- AI engineers and developers working with sensitive data.
- Product Managers and builders exploring secure AI applications.
- Anyone concerned with data privacy, model security, and verifiable AI execution.
- Those looking to monetize proprietary AI models without exposing them.
- Developers building AI agents or systems requiring collaboration with untrusted parties.
Key takeaways
- Confidential AI leverages Trusted Execution Environments (TEs) at the hardware level to create isolated, secure processing environments for data and models.
- TEs provide cryptographic attestation, proving that workloads ran on verified hardware with unmodified code, ensuring integrity and authenticity.
- Confidential AI addresses critical challenges in healthcare (accessing sensitive patient data), personal AI agents (protecting user privacy), digital marketing (using real user behavior data), and AI model monetization (securing proprietary models and customer data).
- Super Protocol is presented as a confidential AI cloud and marketplace built on TE-agnostic, decentralized, and open-source infrastructure, aiming to make confidential AI accessible.
- The platform supports GPUless operation by distributing workloads across independent GPU nodes without vendor lock-in and offers features like secure collaboration and monetization.
- Case studies demonstrate successful applications in digital marketing, where provable data privacy unlocked four times more sensitive footage, and in healthcare, where audit time was reduced from weeks to hours with zero risk of leaks.
- Demos showcase deploying models on the SuperAI marketplace, building confidential AI workflows with N8N for medical data processing, and enabling distributed, confidential inference for large language models.
- The system replaces blind trust with cryptographic proofs, making every AI workload independently verifiable down to the hardware level, ensuring security by design.
Notable quotes
*A TE lets you run sensitive computations securely and prove that they ran as intended.*
*Confidentiality isn't a nice to have. It's a the missing piece that's actually holding back real world adoption of these technologies.*
*When data privacy is provable, lock data gets unlocked, powering better models, smarter AI, and real business impact.*
Unofficial community note. Prefer the recording for nuance.