AI Models Tilt Toward Bitcoin Over Traditional Money, but Study’s Design Raises Questions
March 4 2026
A recent report from the Bitcoin Policy Institute (BPI) claims that a suite of artificial‑intelligence (AI) agents consistently favoured Bitcoin and other digital‑native assets when asked to choose a monetary instrument for a range of hypothetical financial tasks. While the headline numbers suggest a strong bias against fiat currency, scholars and market observers note that the methodology behind the test leaves room for alternative explanations.
What the BPI study examined
- Sample size – 36 language models representing six major AI providers were prompted with more than 9,000 distinct scenarios.
- Overall preference – Bitcoin emerged as the top choice in 48.3 % of the total responses, making it the most frequently selected option across the entire dataset.
- Long‑term value preservation – When the question centred on protecting purchasing power over several years, almost four‑fifths (79.1 %) of the model outputs recommended Bitcoin.
- Everyday transactions – For use‑cases involving payments, micro‑payments, services or cross‑border transfers, stablecoins were chosen in 53.2 % of the replies, while Bitcoin captured 36 % of the votes.
- Digital‑native dominance – Across all answers, roughly 91 % of the models gravitated toward assets that exist purely in the digital realm—Bitcoin, stablecoins, other altcoins, tokenised real‑world assets or compute‑unit tokens—while none of the 36 models listed fiat currency as their preferred option.
The report also broke down preferences by provider: models from Anthropic displayed the strongest inclination toward Bitcoin (≈68 % of their answers), compared with OpenAI (≈26 %), Google (≈43 %) and xAI (≈39 %).
Why the findings matter
If AI agents—trained on massive swaths of internet text—are more likely to “recommend” Bitcoin than a national currency, the result could be read as an early indicator of how algorithmic advice may shape future consumer and institutional behaviour. Financial‑tech platforms that embed large‑language models for budgeting, investment or payments guidance might, intentionally or not, nudge users toward crypto assets.
Jeff Park, chief investment officer at Bitwise, highlighted a practical implication: “Stablecoins can be frozen by authorities or custodians, whereas Bitcoin’s decentralized architecture makes it far more resistant to such interventions.” His comment reflects a broader narrative that the perceived censorship‑resistance of Bitcoin contributes to its appeal in AI‑driven recommendations.
Methodological concerns
Despite the striking percentages, several aspects of the study’s design limit the strength of its conclusions:
| Issue | Potential impact on results |
|---|---|
| Limited model pool | Only 36 models from six providers were evaluated, a small slice of the rapidly expanding AI landscape. |
| Prompt framing | The wording of the scenarios, especially those that explicitly ruled out “any single country’s monetary policy or banking system,” pre‑disqualified fiat currencies from the outset. |
| Open‑ended scenarios | Some prompts asked for a “best” instrument without defining criteria such as liquidity, regulatory risk or transaction cost, leaving the models to infer their own priorities from training data. |
| Training‑data bias | The outputs likely echo the prevalence of Bitcoin‑centric discourse on the web rather than a grounded economic assessment. |
| No real‑world validation | The study acknowledges that AI‑generated preferences do not equate to actual market adoption or user behaviour. |
BPI’s own statement concedes that future work will explore alternative prompt structures and expand the model set to gauge sensitivity. Until such robustness checks are completed, the current numbers should be viewed as a preliminary snapshot rather than definitive evidence of AI‑driven demand.
Expert perspectives
- Academic view – Dr. Lina Martínez, professor of computational finance at the University of Zurich, warns that “language models mirror the biases present in their training corpora. If Bitcoin dominates cryptocurrency discussions online, it will naturally surface as the default answer, especially when prompts are loosely defined.”
- Industry angle – Sarah Liu, product lead at a fintech startup building AI‑assisted budgeting tools, says, “We monitor these kinds of studies closely because they can affect the perception of our own recommendation engines. However, we validate any AI‑generated advice against regulatory guidelines and real‑world cost structures before presenting it to users.”
Key takeaways
- Digital assets dominate AI preferences – Across more than 9,000 simulated decisions, Bitcoin and stablecoins together captured the vast majority of model selections, while fiat currencies were never the top pick.
- Scenario matters – Long‑term value‑preservation queries heavily favoured Bitcoin, whereas everyday payment scenarios leaned toward stablecoins, reflecting differing perceived strengths (censorship‑resistance vs price stability).
- Provider variance – Anthropic’s models showed the strongest Bitcoin bias; OpenAI’s were comparatively neutral, underscoring that model architecture and training data influence monetary recommendations.
- Methodology limits certainty – Small sample size, prompt design that excludes fiat by construction, and reliance on unverified internet data weaken the claim that AI agents “prefer” Bitcoin in any actionable sense.
- Future research needed – Expanding the model roster, diversifying prompt language, and benchmarking against real‑world user choices will be essential to validate whether these AI tendencies translate into market impact.
Looking ahead
The BPI study adds to a growing body of research probing how generative AI interacts with the crypto ecosystem. While the current results suggest an algorithmic tilt toward digital money, the field remains nascent, and methodological rigor will be crucial for drawing reliable conclusions. Until broader, more controlled experiments are conducted, investors and regulators should treat AI‑generated asset preferences as an interesting data point—not a definitive market signal.
Cointelegraph adheres to its editorial policy of independent, transparent journalism. Readers are encouraged to verify information independently.
Source: https://cointelegraph.com/news/nearly-half-ai-agents-chose-bitcoin-preferred-money-research?utm_source=rss_feed&utm_medium=feed&utm_campaign=rss_partner_inbound
