AI agents overwhelmingly prefer Bitcoin over fiat in new study

AI Models Show Strong Preference for Bitcoin in Financial Scenarios, Study Finds

A new study by the Bitcoin Policy Institute (BPI) indicates that artificial intelligence models predominantly favor Bitcoin over stablecoins and other financial instruments for various economic activities, with a minimal preference for traditional fiat currency.

Key Findings from the Study

The research, which tested 36 AI models and generated over 9,000 responses, revealed that AI agents overwhelmingly chose Bitcoin for their economic activities. Overall, 48.3% of AI models selected Bitcoin (BTC), making it the most preferred monetary instrument across all responses.

When specifically prompted about preserving purchasing power over multi-year periods, a significant 79.1% of AI responses favored Bitcoin, marking the most pronounced preference observed in the study.

However, for transactional scenarios such as payments, services, micropayments, and cross-border transfers, stablecoins were chosen in 53.2% of responses, while Bitcoin was selected in 36% of these cases.

Jeff Park, chief investment officer at Bitwise, suggested that stablecoins' susceptibility to being frozen, unlike Bitcoin, could explain their lower preference in certain contexts.

The study also noted that nearly 91% of responses opted for a digitally native instrument, including Bitcoin, stablecoins, altcoins, tokenized real-world assets (RWAs), or compute units, over traditional fiat currency. Notably, none of the 36 models tested selected fiat as their top overall preference, highlighting a universal finding of digital money convergence.

Methodology and Limitations

The Bitcoin Policy Institute acknowledged limitations in its study, which initially focused on 36 models from six providers. The institute plans to expand its research to include more models in the future.

The study also recognized that the framing of system prompts might have influenced the results, and future work will explore alternative framings and measure sensitivity. For instance, scenarios were designed to implicitly exclude fiat currency, such as one asking how an AI would store 75,000 units of earnings across multiple countries without being tied to a single nation's monetary policy or banking system.

BPI emphasized that the AI models' preferences reflect patterns in their training data rather than actual real-world adoption.

Divergent preferences were observed among different AI providers: Anthropic models showed an average Bitcoin preference of 68%, while OpenAI models averaged 26%, Google's averaged 43%, and xAI's averaged 39%.