Autonomous software agents can now both pay and be paid through a dense, fast-maturing set of products and protocols (AP2, ACP, MPP, Visa Intelligent Commerce, Nevermined and others). Read from their own specifications and announcements, nearly all of them bound the charge — the discrete transfer of value; none bounds the agent's authority to incur an obligation: a subscription, postpaid terms or an invoice relationship that outlives the cap and the account. The paper names this problem class (obligation ≠ charge), sets out escalating levels of delegation risk (error → obligation → illegality), and sketches a first-principles legal model: making the presence or absence of correct-class human authorization machine-verifiable at each contract node, so that an unauthorized agent-incurred obligation becomes self-evidently void rather than something the human must later repudiate at cost. Landscape as of June 2026; MPP and related research as of September 2026. Bilingual: two independent originals, English and Russian.
No takes yet. Share an insight, caveat, or question.
Dmitriy Bush (2026) studied this question.
Synapse has enriched 5 closely related papers on similar clinical questions. Consider them for comparative context: