Answers are grounded in this paper and cite their sources.
When a software agent spends a person's money, someone must be able to answer, later and to a third party: what exactly did the owner approve, under what limits, and what happened next? This paper describes Sealgraf, a system running in production since 2 November 2026, that removes the trust step for the approval half of that problem. An owner sets machine-checkable limits in advance (a mandate); an agent checks each proposed purchase against those limits before spending; purchases above an owner-set line are decided by the owner personally, on their own authenticated surface. Approval mints a single-use token bound to the hash of the exact action approved, and the outcome is a sealed, owner-held receipt in which the approval is proven and the purchase is reported — kept as separate grades of fact.
No takes yet. Share an insight, caveat, or question.
Sai Arun Gunuganti (2026) studied this question.
Synapse has enriched 8 closely related papers on similar clinical questions. Consider them for comparative context: