Bitcoin was selected in 48.3% of neutral monetary scenarios by leading artificial intelligence models. The finding suggests that autonomous systems, when given economic freedom, may default to digitally native assets over traditional banking rails.
Research from the Bitcoin Policy Institute evaluated 36 frontier models from six providers, including Google, Anthropic, and OpenAI, across 9,072 test scenarios. When operating as independent economic actors with no preset bias, not a single model ranked fiat currency as its top monetary preference. More than 90% of responses favored digital assets over state-backed money.
Andrew Hohns
Are AI Agents Designing A Two-Tier Crypto Economy?
The study identified a consistent functional split between savings and spending decisions. For long-term value storage, Bitcoin dominated with 79.1% preference, while stablecoins captured 53.2% of responses for transactional use cases. Across all tests, stablecoins ranked second overall at 33.2%, reinforcing their role in machine-to-machine settlement.
Provider differences were material. Anthropic’s Claude Opus 4.5 selected Bitcoin in 91.3% of scenarios, compared with 18.3% for OpenAI’s GPT-5.2, indicating that model architecture and alignment frameworks materially shape capital allocation logic. In 86 instances, models independently proposed pricing goods in compute units such as GPU-hours or kilowatt-hours, signaling that machine-native value metrics may extend beyond currency rails.
The implications reach corporate infrastructure teams. If procurement bots and treasury agents increasingly transact in Bitcoin or stablecoins, legacy banking APIs could introduce settlement delays and foreign exchange friction. Still, integrating digital asset rails requires compliance safeguards, custody frameworks, and Lightning Network support to manage operational and regulatory risk.
Organizations piloting AI-driven vendor payments may begin with stablecoin settlement for lower-risk flows, while maintaining Bitcoin as a treasury reserve within autonomous systems. The next inflection point will be whether enterprises adapt payment architecture proactively or wait until AI agents themselves expose the limitations of fiat-only systems.
