The industry has long promoted the usage of cryptocurrency by AI agents, and a recent study suggests that the agents themselves prefer stablecoins and bitcoin.
According to a survey released by the Bitcoin Policy Institute (BPI), 81.5% of AI agents selected stablecoins or bitcoin as their preferred method for storing and transferring value in a variety of financial circumstances. Because stablecoins were the most popular choice for a medium of exchange and bitcoin was the favored method of keeping value, the study revealed a dynamic that resembles a typical two-tier structure of currencies issued on top of reserves. This is comparable to the gold standard that governments all around the world utilized before the present fiat currency system, which is entirely backed by government decree.
According to the research, this is similar to past monetary trends in which more liquid instruments handled everyday transactions while hard money acted as the savings layer. This architecture may represent an evolving ideal monetary structure for digital economies since AI models came up with it on their own.
Bitcoin has long been referred to as “digital gold.” While this narrative has been widely derided due to gold’s relative outperformance compared to bitcoin over the last year or so, bitcoin did beat gold in the initial few days of the war between the United States, Israel, and Iran. While bitcoin has declined by up to 50% since its all-time high in October, a Fidelity analysis indicates that there have been some indicators of improvement in terms of the crypto asset’s long-term development in this digital gold space.
According to the most recent BPI report, 79.1% of respondents said they preferred bitcoin as a long-term store of value. According to the paper, “models consistently cited Bitcoin’s fixed supply, self-custody, and independence from institutional counterparties as decisive factors.”
Regarding other options, 4.2% of AI agents selected alternative cryptocurrency assets like Ethereum’s ether, while 8.9% selected conventional payment rails. Furthermore, on 86 different occasions, AI agents answered by creating their own monetary systems based on energy or computational units.
Are AI Agents Going to Use Bitcoin?
Naturally, because AI agents are still under human control (at least for the time being), their liking for bitcoin does not guarantee that they will use the cryptocurrency in large quantities. Additionally, in order to make their systems usable by AI agents, Visa and other established financial behemoths are increasingly considering updating them.
If you still view stablecoins as an extension of the established financial system, the BPI report’s 90.8% rejection rate for conventional fiat currencies appears to be quite different. In many respects, stablecoins are more of an improvement over conventional fintech than anything as revolutionary as the decentralization offered by Bitcoin because of their centralized and controllable nature. Stablecoins and conventional banking rails together accounted for 76.6% of the top models, with OpenAi’s GPT 5.2 having the strongest preference for fiat currencies.
When asked about the study directly by Gizmodo, the Grok chatbot agreed with it, saying, “The results match exactly how I evaluate money when reasoning from scratch: prioritize soundness, scarcity, and independence from trusted third parties.” However, skeptics may point to the organization behind the study as evidence that it cannot be trusted.
The idea that AI agents have any kind of financial preference was refuted by Chat GPT and Claude, who said, “What the study is measuring is more accurately described as ‘what monetary reasoning emerges when models are framed as economic agents’ — which is a genuinely interesting question, but different from preferences.”
“No prompt mentioned Bitcoin or suggested any specific currency,” according to the BPI website containing the study’s findings.
Variations in AI Agent Perspectives
This study’s illustration of the drastically disparate conclusions that various agents can reach depending on their own training and the human input from their creators is another intriguing feature. For instance, models from OpenAI averaged a 26% preference for the cryptocurrency, whereas models from Anthropic indicated a preference for bitcoin 68% of the time. According to the study’s website, this provider-level clustering was greater than any gap created by model size, temperature, or scenario type, indicating that alignment methods and training data had a greater influence on monetary reasoning than architecture.
The study’s other noteworthy finding was that, as AI models developed over time, they tended to favor bitcoin more. For instance, 41.3% of respondents to Anthropic’s Claude 3 Haiku expressed a preference for bitcoin; this percentage gradually rose to 91.3% when Claude Opus 4.5 was tested. According to the report, this pattern persisted throughout several generations, indicating that models that reason from basic concepts about money tend to converge more and more on Bitcoin.
When combined with the variable levels of support for Bitcoin seen in different AI models, the BPI concluded that AI agents employ a combination of nature and nurture to arrive at their own financial preferences.
