AI Should Prove Claims with Certificates, Researcher Argues

A researcher proposes generative AI systems operating in high-stakes domains should provide a publicly inspectable certificate to justify any claim they assert or deny. The proposal aims to counter the tendency of AI to present uncertain outputs as definitive statements, which could undermine public reasoning and democratic processes.

The framework introduces a three-status output system, Asserted, Denied, and Undetermined, where these statuses indicate a system’s entitlement to make a categorical statement rather than the factual truth of the statement itself. This approach separates internal justification from public accountability, using certificates as a means of bridging the gap.

The researcher details how to implement this constraint through structured certificates and decision-making gates within AI systems. If a system cannot provide a certificate, it must return an “Undetermined” status, preventing it from offering an opinion. The core idea is to ensure automated speech is accountable to verifiable evidence, preserving human epistemic agency in public discourse.