SPAR Mentoring

In Spring 2026, I mentored two SPAR projects:

Proving Model Equality in Zero-Knowledge. When an AI model is deployed as a service, users have no guarantee that it is the same model that was audited or certified. The project explored zero-knowledge proof systems and challenge-response protocols that let a model operator prove that the deployed model matches a reference model, without revealing the weights.

Proving Safety Properties of Guardrail Models. Guardrail models such as constitutional classifiers enforce safety policies, but nothing formally guarantees they satisfy the properties they claim. The project applied formal verification to prove safety properties of guardrail models, rather than relying on empirical testing alone. Joint with Luca Arnaboldi.