AI as Trusted Third Parties for Secure Computations

A recent article explores the idea that AIs can act as trusted third parties, replacing human intermediaries or cryptographic protocols for secure computations. The document, titled "Trusted Machine Learning Models Unlock Private Inference for Problems Currently Infeasible with Cryptography," proposes Trusted Capable Model Environments (TCME) as an alternative for secure computations, enabling private inferences where traditional cryptographic solutions are currently inefficient. TCMEs use capable machine learning models with input/output constraints and explicit control of information flow. The article describes several use cases, including classical cryptographic problems solved by TCMEs. For example, Alice and Bob can use a trusted AI model to determine who has the higher income without revealing their respective incomes. This approach allows for the auditing of AI processing and the removal of its knowledge after use.