BlackRock: AI agents need machine-native money

BlackRock published a research paper arguing that AI agents require “machine-native money” and that stablecoins are the leading candidate for instant on-chain payments. The paper states “stablecoins, native cryptoassets, and other on-chain assets can serve as machine-native instruments” and cites a study showing AI models often preferred Bitcoin as a store of value.

The paper outlines practical limits of current payment rails. Autonomous software cannot open bank accounts or payment cards on its own. Merchant fees make sub-dollar payments uneconomic and Automated Clearing House transfers can take up to a business day to settle. The report contrasts those limits with blockchains, which settle near real time and operate around the clock.

BlackRock sketches a two-tier system for machine money: stablecoins for immediate spending and native cryptoassets such as Bitcoin for longer-term savings. The paper highlights payment protocols designed for machines, including one that revives the HTTP 402 “Payment Required” code to enable instant data and service payments by agents. The report also lists alternative payment rails from large technology and payments firms that would use bank money. It notes that live agent payment volumes remain small.

The paper presents scale figures for stablecoins to support the case. BlackRock reports circulating stablecoin supply above $300 billion. It says adjusted stablecoin volume exceeded $11 trillion in 2025, a figure it compares with a roughly $11.2 trillion figure for a major card network and a larger $16.7 trillion figure in another comparison. The report adds that stablecoins moved about $8.5 trillion in the first half of 2026. BlackRock launched a money market fund earlier in the year aimed at stablecoin issuers looking to park reserves.

For the question of where machines should save value, the paper cites a February 2026 study by the Bitcoin Policy Institute. That study tested 36 frontier language models from several vendors and collected 9,072 answers. When asked where to store value, the models chose Bitcoin 79.1% of the time; when asked what to use for spending, they chose stablecoins 53.2% of the time. Bank-based money received under 9% overall. The study found variation by model vendor, with some models favoring Bitcoin more than others.

The report repeatedly uses conditional language and includes a disclaimer that it is not a forecast. It describes the ideas as preliminary and notes that current on-chain agent payment activity is limited.