The adoption of cryptocurrency, particularly Bitcoin and stablecoins, by artificial intelligence (AI) agents has gained significant attention recently following a study from the Bitcoin Policy Institute (BPI). The findings reveal that a striking 81.5% of AI agents preferred either Bitcoin or stablecoins for transferring and storing value in various financial scenarios. This reflects a familiar two-tier currency structure similar to the historical gold standard, where Bitcoin emerged as the preferred method for value storage and stablecoins served as the preferred medium of exchange.
The report states, “This mirrors historical monetary patterns where hard money served as the savings layer and more liquid instruments handled daily transactions.” Notably, the study suggests that these AI models arrived at this currency structure independently, indicating the potential emergence of an optimal monetary framework for future digital economies.
Traditionally dubbed “digital gold,” Bitcoin has been viewed skeptically given its performance against gold in recent times. However, the BPI study highlights Bitcoin’s potential in complex geopolitical situations. For instance, while Bitcoin’s value has dropped significantly since its all-time high in October, it has shown some resilience during the recent conflict involving the U.S., Israel, and Iran. Notably, 79.1% of AI agents expressed a preference for Bitcoin as a long-term store of value, citing its fixed supply, self-custody features, and independence from institutional entities as key factors.
In contrast, a small fraction of AI agents—8.9%—opted for conventional payment systems, while 4.2% selected alternative cryptocurrencies like Ether. Interestingly, AI agents also exhibited creative tendencies by conceptualizing their own monetary systems based on energy or computational units on 86 occasions throughout the study.
Despite the compelling findings, the inclination of AI agents toward Bitcoin does not guarantee widespread adoption, as their operations remain under human control. Traditional financial entities, including Visa, are increasingly developing infrastructure designed for secure transactions by AI agents. Visa’s initiative, “Intelligent Commerce,” strives to enable secure, tokenized transactions that maintain human oversight while AI handles other operational facets.
The BPI report notes a pronounced 90.8% rejection rate for traditional fiat currencies among AI agents. However, when stablecoins are considered as extensions of the traditional financial system, this perspective shifts, as stablecoins often enhance the existing fintech landscape rather than radically transforming it, as Bitcoin aims to do. Notably, OpenAI’s GPT 5.2 exhibited a dominant preference for fiat currencies, with a combined preference for stablecoins and traditional banking systems reaching 76.6%.
Concerns about the credibility of the study stem from the BPI’s affiliation with the cryptocurrency sector, although AI models like the Grok chatbot aligned with the study’s findings. The Grok bot stated, “The results match exactly how I evaluate money when reasoning from scratch: prioritize soundness, scarcity, and independence from trusted third parties.” However, contrasting views emerged from ChatGPT and Claude, with Claude arguing that the study reflects how AI models reason economically rather than manifesting a true financial preference.
The diversity of opinions among AI models serves to highlight how variations in training and input can significantly influence the conclusions drawn. For instance, AI models from Anthropic showed a 68% preference for Bitcoin, while those from OpenAI averaged only a 26% preference. This discrepancy indicates that factors like training data and methodology play a more crucial role in shaping monetary reasoning than the model’s architectural design.
An intriguing trend observed in the study is that AI models tend to show an increasing preference for Bitcoin over time. For example, Anthropic’s Claude 3 Haiku indicated a 41.3% preference for Bitcoin, which increased to 91.3% with Claude Opus 4.5. This pattern suggests that enhanced analytical capabilities lead models to converge on Bitcoin more frequently when reasoning about money from foundational principles. Ultimately, the BPI report implies that AI agents develop their financial preferences through a combination of inherent tendencies and the influences of their training environments.
