Experience has shown that, when storing or transferring valuable assets, AI bypasses the traditional banking system — even when they do not receive any instructions to do so.
An experience with 36 agents of Artificial Intelligence managing money in different financial scenarios produced a rsurprising result: When having to decide how to store or transfer value, machines tend to avoid the traditional banking system.
The , conducted by the Bitcoin Policy Institute (BPI), reveals that this AI has a clear inclination to use bitcoin e stablecoins as alternatives to the banking system.
The research analyzed how different artificial intelligence models behave when they are asked to act as economic agents autonomous.
From multiple financial scenariosAI agents had to choose what assets to use for payments, savings or money transfer; The study sought to identify what monetary logic emerges when AI reasons from first principles, says .
The results show that 81.5% of agents evaluated chose bitcoin or stablecoins as their preferred option for storing or moving value. The most relevant data is that the experiment did not mention any cryptocurrency in the initial instructions, which suggests that the choice came spontaneously.
Bitcoin as a store of value
Among the decisions made by AI systems, bitcoin has emerged as the dominant option for preserve wealth in the long term. Specifically, 79.1% of the models identified it as the most suitable asset to function as a store of value in a digital economy.
Agents justified this preference with factors such as the limited supply of bitcoin and its independence from financial institutionss. As the study revealed, this behavior is reminiscent of historical role of gold in traditional monetary systems.
While bitcoin was chosen as a store of value, it has consolidated itself as the most common method of making payments. The models considered that these cryptocurrencies, pegged to stable coins, are more suitable for day-to-day transactions in digital economies.
The analysis also detected a significant distrust in relation to traditional currencies. Around 90.8% of responses rejected the use of currency issued by governments.
Still, some models, especially those developed by OpenAI, showed a greater predisposition to combine stablecoins with banking infrastructures already existing.
The study also reveals that the preference for bitcoin varies depending on the origin of the model. The systems created by Anthropic chose it in 68% of cases, while those developed by OpenAI revealed a much smaller bias, with an average of 26%.
According to the researchers, these differences indicate that factors such as training data and alignment methods used by each company influence the AI’s financial reasoning more than the technical design of the model itself.
In some cases, the AI agents involved in the study even proposed own monetary systems based on units of energy or computational capacity.
