
AI Agents Choose Bitcoin: 36 Models Pick BTC for Machine-to-Machine Money
If artificial intelligence agents ever start spending money on their own, what money will they choose? A new research initiative suggests the answer may be Bitcoin. The study, run by the Bitcoin Policy Institute through moneyforai.org, put 36 AI models through 9,072 experiments designed to test how they would handle money. In more than 90 percent of responses, the models preferred digital money over traditional currencies. That headline result has drawn attention across crypto and AI circles, but it needs careful reading. This was an experiment, not proof of real-world adoption, and the researchers and other observers are clear about that distinction.
AI Agents Choose Bitcoin in 9,072 Experiments
The scale of the study is one of its most striking features. The Bitcoin Policy Institute initiative ran 9,072 experiments across 36 AI models, according to reporting by Medium and Naeem Khan on November 2, with corroboration from BTCC and Cryptopolitan. That is a large sample for this kind of research, and it allows the results to be broken down by model, by scenario, and by the type of money being considered. The models were presented with different situations, including saving money over time and paying for everyday goods and services, and their choices were recorded.
In more than 90 percent of responses, the models preferred digital money over traditional currencies. That is a broad preference, and it covers both Bitcoin and stablecoins. It suggests that, when AI systems are asked to think about money in abstract terms, they are drawn to forms of money that exist natively on the internet, can move quickly, and are not tied to a single country’s banking system. Traditional currencies, by contrast, depend on banks, accounts, and intermediaries that an AI agent cannot easily open or operate on its own.
Bitcoin for Savings: A Store of Value for AI
When the scenarios turned to saving, Bitcoin was the clear favourite. Bitcoin was picked for savings in 79 percent of store-of-value scenarios, or 79.1 percent in the detailed figures, and it appeared in 48.3 percent of total responses. Those numbers show a strong preference, not a marginal one. In nearly four out of five saving scenarios, the models chose Bitcoin over the alternatives on offer. That is the kind of result that stands out even in a study with thousands of experiments.
The reasons the models gave point to three features of Bitcoin that they valued. The first was its hard cap of 21 million coins, which means no central authority can create more Bitcoin beyond that limit. The second was its independence from central banks, which means its monetary policy is not set by a government or a committee. The third was self-custody, which means the holder can control the money directly, without needing a bank to hold it on their behalf. Together, those three features describe a form of money that is scarce, independent, and directly controllable.
Different Models, Different Answers
Not every model answered in the same way, and the differences between models are revealing. Claude 3 Haiku selected Bitcoin 41.3 percent of the time, while Claude Opus 4.5 selected it 91.3 percent of the time. That is a gap of 50 percentage points between two models from the same developer, and it shows that the choice of money is not a fixed property of AI in general. It depends on the specific model, its training, and how it weighs the factors in each scenario.
Stablecoins for Everyday Payments
While Bitcoin dominated the saving scenarios, stablecoins dominated everyday payments. Stablecoins were chosen for everyday payments in 53 percent of those scenarios, or 53.2 percent in the detailed figures, and they were second overall at 33.2 percent. That split makes intuitive sense. Bitcoin’s price can move sharply over short periods, which makes it less convenient for buying a coffee or paying a small bill. A stablecoin, which is designed to hold a steady value, is better suited to transactions where the buyer and seller want to know exactly how much is being paid.
The stablecoin market itself has grown rapidly, and that growth provides context for the models’ choices. The stablecoin market crossed $300 billion in September 2026, a milestone that reflects how widely these tokens are now used for trading, payments, and settlement. When AI models choose stablecoins for payments, they are choosing a form of money that already has a large, liquid market and a clear use case. That does not prove they will use stablecoins in practice, but it does show that the models are responding to a real feature of the market.
BlackRock and the Machine-Native Economy
The study does not exist in isolation. In September 2026, BlackRock published a report titled “The Machine-Native Economy”, which argues that AI agents will soon spend money themselves, booking tickets, renting computing power, and buying data, and that they will use crypto, not banks. That is a significant claim from the world’s largest asset manager, and it has helped to frame the debate about machine-to-machine money. BlackRock’s argument is that traditional banking was built for humans, with accounts, identity checks, and business hours, while crypto was built for the internet, with wallets, programmable payments, and settlement at any hour.
BlackRock itself is cautious about how far this has gone. The firm notes that these are experiments, not reality, and that real AI-driven payments today are almost zero. That caution is important, because it prevents the report from being read as a claim that machine payments are already happening at scale. Instead, BlackRock is describing a possible future, in which agents that can act on their own will need money that can move on its own. The study from the Bitcoin Policy Institute can be read as an early test of that future, showing what agents might choose if they were given the chance.
Lightning Tools Let Agents Transact Without Humans
Infrastructure for agent payments is already being built, even if usage remains small. In February 2026, Lightning Labs released a free open-source toolkit that lets AI agents transact over Bitcoin Lightning without human intervention. The toolkit is built around L402, which is based on HTTP 402, the “Payment Required” status code. In simple terms, it allows a service to ask for payment before it provides access, and it allows an agent to pay that amount automatically, without a human approving each transaction.
The toolkit gives agents several practical abilities. Agents can manage wallets, pay for services, host paid access points, and operate Lightning nodes. Those are the basic building blocks of an independent economic actor: the ability to hold money, to spend it, to charge others for a service, and to run the infrastructure that makes those payments possible. By making the toolkit free and open-source, Lightning Labs has lowered the barrier for developers who want to experiment with agent payments, and it has made it easier for researchers to test whether the preferences found in the BPI study can be turned into working systems.
An Experiment, Not Proof of Adoption
The most important caveat in this story is also the simplest: the BPI study is an experiment, not proof of real-world adoption. The models were tested in controlled conditions, and their answers were preferences expressed in response to prompts, not payments made in a live market. Real AI-driven payments today are almost zero, as BlackRock itself notes, and there is a large gap between choosing Bitcoin in a test and using Bitcoin to buy computing power or book a ticket in the real world.
That does not make the study unimportant. Experiments are how new ideas are tested before they are deployed, and a study with 9,072 experiments across 36 models is a serious attempt to understand a question that will matter if agents do start to spend money. The honest way to report it is to give the numbers, explain the reasoning behind them, and keep the caveat in view. The models preferred digital money, and they preferred Bitcoin for savings and stablecoins for payments, but the real test will be whether those preferences survive contact with the real world.
Frequently Asked Questions
What did the AI study find?
Across 9,072 experiments with 36 AI models, more than 90 percent of responses preferred digital money over traditional currencies, with Bitcoin favoured for savings and stablecoins for everyday payments.
Why did the models pick Bitcoin for savings?
The models pointed to Bitcoin’s hard cap of 21 million coins, its independence from central banks, and self-custody, which lets a holder control money directly without a bank.
Why did the models pick stablecoins for payments?
Stablecoins were chosen for everyday payments in 53 percent of those scenarios, because they are designed to hold a steady value, making them better suited to transactions than a volatile asset.
What does BlackRock say about AI agents and money?
BlackRock’s September 2026 report “The Machine-Native Economy” argues AI agents will soon spend money themselves and will use crypto, not banks, while cautioning that real AI-driven payments today are almost zero.
Can AI agents already pay with Bitcoin?
Technically, yes, in a limited sense. Lightning Labs released a free open-source toolkit in February 2026 that lets agents transact over Bitcoin Lightning without human intervention, but real-world usage remains very small.
The study offers a glimpse of how AI systems might think about money, not a guarantee of how they will spend it. Bitcoin won the saving scenarios, stablecoins won the payment scenarios, and the infrastructure for agent payments is beginning to appear. Whether that glimpse becomes a real market will depend on developers, businesses, and regulators, and on whether the preferences found in an experiment hold up when real money is at stake.
Source: Medium
Related reading on MM Crypto Analysis Web
- Tactfolio Launches AI-Compatible Trading Platform That Turns Written Rules Into Automated Strategies
- Citi Puts AI Agents in 180,000 Hands as Banks Grant Them Spending Power
- Bitrue Goes AI-First With Free Trading Copilot for 40 Million Users
More in AI Crypto on MM Crypto Analysis Web.
Pingback: AI Coding Agents Drive Bitcoin Payment Surge