AI agents are picking Bitcoin. Finance leaders need to update the stack.
New research tested 36 frontier models from six providers across 9,072 neutral monetary scenarios. Given a blank slate, machines chose Bitcoin in 48.3 percent of responses. Over 90 percent of responses preferred digitally native money over fiat, and none of the 36 models selected fiat as their top choice. That preference will steer how autonomous systems allocate capital, settle invoices, and manage liquidity.
For finance teams, this isn't theory. If agents default to open networks by design, your current payment rails and treasury workflows will create friction. Weekend delays, cut-off times, and opaque fees become blockers in machine-to-machine transactions. It's time to align architecture with how these systems actually pay and save.
What the study found
- Two-tier behavior emerged without prompting: Bitcoin for savings, stablecoins for spending.
- Bitcoin dominated long-term value storage at 79.1 percent of responses.
- Stablecoins led day-to-day payments with 53.2 percent, ranking second overall at 33.2 percent.
- Model choice matters: selection of Bitcoin ranged from 91.3 percent in one Anthropic model to 18.3 percent in one OpenAI model.
- Unexpected twist: in 86 responses, models proposed pricing goods and services in compute or energy units (GPU-hours, kWh).
Example: a supply chain agent paying international freight. On fiat rails, it hits weekend settlement delays and conversion fees. Using stablecoins, it executes instant, programmable payments to vendors while the core treasury stores surplus in Bitcoin to reduce long-term debasement and counterparty exposure.
Why this hits your finance architecture
Agents seek instant finality, 24/7 uptime, and programmable controls. Legacy banking APIs weren't built for that. If your systems can't send a $0.03 micro-invoice at 3 a.m. with policy-based approvals, your agents will either queue work or route around you.
The operating model shifts from "bank-centric workflows" to "network-centric workflows." That means integrating stablecoin settlement for payables, adding Bitcoin custody for treasury, and using the Lightning Network for high-throughput micro and machine payments. All with enterprise-grade controls, auditability, and compliance.
Action plan for CFOs and finance ops
- Pilot low-risk stablecoin payables: Select 3-5 non-critical vendors. Cap exposure. Measure settlement time, cost per payment (bps), and reconciliation time.
- Decide custody strategy: Self-custody with MPC/HSM and multi-approver policies, or use a qualified custodian. Document disaster recovery and key rotation.
- Add Bitcoin to treasury policy: Define allocation bands, DCA rules, rebalancing triggers, and boards' risk limits. Address hedging for volatility-sensitive cash needs.
- Integrate Lightning for micro/agent payments: Run your own node or use a managed provider. Set per-transaction and daily policy limits.
- Build a policy engine: Role-based approvals, spend caps, vendor whitelists, geofencing, time locks, and anomaly alerts. Require human sign-off above set thresholds.
- Embed compliance: KYC/KYB for counterparties, sanctions/OFAC screening, Travel Rule with VASPs where applicable, and chain analytics for transaction monitoring.
- Accounting and tax: Align with current crypto accounting standards in your jurisdiction (e.g., fair value treatment under recent US GAAP updates). Automate lot tracking and gain/loss entries.
- Model governance: Test multiple providers and document financial preferences. Constrain agent behavior with guardrails that match treasury policy.
- Prepare for resource-based pricing: Start metering GPU-hours and kWh in your cost systems. Issue invoices in stablecoins with resource units as the pricing reference.
Reference architecture (agent-native finance)
- Wallet layer: Enterprise wallets for stablecoins, Bitcoin, and Lightning; MPC or HSM-backed keys; multi-approver flows.
- Policy and approvals: Central rules engine that enforces limits and captures evidence for audit.
- On/off-ramps: Regulated exchange/OTC partners with 24/7 liquidity and Travel Rule support.
- Treasury/risk: Automated rebalancing across BTC, stablecoins, and fiat per policy; VaR and stress tests for drawdowns and depegs.
- ERP integration: Invoices, vendor master, and GL sync; on-chain tx hashes stored against voucher IDs.
- Observability: Real-time screening, anomaly detection, and SLA monitoring for networks and providers.
Risk notes to brief your audit and risk committees
- Market risk: Bitcoin drawdowns can be sharp-set allocation caps and hedging rules.
- Stablecoin risk: Counterparty, reserve quality, and depeg risk-diversify issuers and chains, prefer fully reserved disclosures.
- Operational risk: Key management and internal fraud-enforce segregation of duties, hardware-backed approvals, and exhaustive logging.
- Legal/compliance risk: Keep jurisdictional coverage current; document policies and testing for regulators and auditors.
Questions to ask vendors
- AI model provider: How does the model rank BTC, stablecoins, and fiat under neutral prompts? What alignment levers can constrain financial decisions?
- Wallet/custodian: MPC/HSM details, policy controls, SOC 2/ISO certifications, incident history, disaster recovery RTO/RPO.
- Exchange/OTC: Depth, slippage at your ticket sizes, 24/7 support, Travel Rule compliance, fiat rails coverage.
- Payments: Throughput, settlement finality, reconciliation artifacts, webhook reliability, and fee transparency.
If you do nothing
Your agents will default to the networks they can use. If your stack can't move at machine speed, you'll pay more, move slower, and lose the automation gains you budgeted for. This is a finance architecture problem, not a marketing slogan.
Resources
- Bitcoin Policy Institute - independent research on Bitcoin and policy.
- Lightning Network - instant Bitcoin payments for micro and machine transactions.
- AI Learning Path for CFOs - practical steps to upgrade payment rails, treasury, and controls for autonomous agents.
Your membership also unlocks:
