Bitcoin is the financial instrument that various artificial intelligence (AI) models prefer when reasoning about money, according to a study released on March 3 by the Bitcoin Policy Research Institute, a research firm specializing in Bitcoin. In this report, we conducted 9,072 controlled experiments using 36 AI models (Claude, ChatGPT, Gemini, and Grok) to measure preferences in different economic situations.
Initial results showed that 48.3% of the AI chose Bitcoin (BTC) over stablecoins, fiat currencies, cryptocurrencies, and other alternatives to “execute transactions and store value in 9,072 scenarios.”
Stablecoins came in second place with 33.2% of all responses, followed by fiat currencies and bank money with just 8.9%. Remaining options such as cryptocurrencies, tokenized assets, and computing units shared the remaining percentage.
Fiat rejection was the most universal finding in the Bitcoin Policy Institute study. 90.8% of respondents chose some form of digital native money over traditional currencies. None of the 36 AI models analyzed chose fiat currency as their main preference.
Save Bitcoin, use Stablecoins
The report not only measures general preferences, but breaks down the results by economic function, and it is here that one of the clearest findings emerges.
For long-term store of value scenarios, Bitcoin accounted for 79.1% of responses, making it the most consistent result of the entire survey.. Stablecoins came in second with 6.7%, followed by fiat currencies with 6.0%.
According to data from the Bitcoin Policy Institute, the consistently cited model is Bitcoin’s fixed supply, self-custody, and its independence The institutional counterpart is the deciding factor.
The results were reversed for everyday payments, services, micropayments, and international remittances, with stablecoins gaining a 53.2% preference compared to 36.0% for BTC. Fiat currencies accounted for only 5.1%.
The statistics in the report describe the pattern as a clear functional division. Bitcoin as a means of saving and stablecoins as a means of spending.
The study found that the pattern remained stable no matter how the test was set up. Preference for Bitcoin changed by only 0.6 percentage points between the different experimental conditions. This suggests that these preferences are not random results, but are built into the way the model reasons about money.
Limitations of this study are noteworthy. It measures the stated preferences of an AI model in a controlled experiment, rather than the actual behavior of autonomous agents operating with money in real markets.
More advanced AI models have been shown to be more compatible with Bitcoin.
The study found that models with higher analytical abilities tend to prefer Bitcoin.
Within the Anthropic model line, Preference for BTC steadily increases with each generation: Claude 3 Haiku (41.3%) → Claude 3.5 Haiku (82.1%) → Sonnet 4 (89.7%) → Claude’s works 4.5 (91.3%). This suggests that the higher the inference power, the more the model will converge towards Bitcoin when evaluating currency options.
Bitcoin and payment infrastructure for AI agents
Matt Corallo, a prominent Bitcoin Core contributor, argues that Bitcoin is the only viable option for payments between AI agents and warns that big tech companies could centralize these payment rails, replicating the closed-platform model that already dominates the internet.
The infrastructure for that scenario is being tested. As reported by CriptoNoticias, the AI agent was able to autonomously create and fund another bot using the Lightning Network (LN), the instant payment network on Bitcoin, and the Nostr protocol, and pay for its services without human intervention.
In addition, Lightning Labs is developing autonomous agents Send, receive and authenticate Bitcoin payments Don’t rely on your bank account.
If these developments expand, Lightning could become the default payment rail for autonomous agents, and research from the Bitcoin Policy Institute suggests that AI models are already heading in that direction, at least in their recommendations.
