## Old Laws for New Machines

## Source details

**Canonical URL:** [Old Laws for New Machines](https://www.imf.org/-/media/files/publications/fandd/article/2026/03/bossone.pdf)

## Other formats

- [Markdown version](/-/media/files/publications/fandd/article/2026/03/bossone.pdf.md)
- [Structured JSON version](/-/media/files/publications/fandd/article/2026/03/bossone.pdf.json)

---

### Asimov’s laws and an ethics hierarchy for AI in finance
- Asimov’s original principles: three rules from the 1940s (avoid harming humans; obey legitimate commands unless causing harm; protect own functioning while respecting higher duties) and the later “Zeroth Law” (safeguard humanity as a whole).
- Reinterpreted hierarchy for finance:
  - First: do no harm to financial stability or consumer integrity.
  - Second: obey institutional mandates without compromising fairness or safety.
  - Third: preserve resilience while remaining accountable.
  - Highest: serve the social contract of money by preserving trust.

### Doing no harm: operational meaning and risks
- AI can misinterpret patterns, inherit historical biases, or reinforce market signals that accelerate volatility.
- Key operational requirements to prevent harm:
  - Rigorous testing.
  - Sound data governance.
  - Explainability as an institutional requirement (current AI models cannot explain their reasoning; they can generate post hoc narratives).
- Macro-financial effects and examples:
  - Algorithms reacting at machine speed can make price adjustments and liquidity shifts sharper and more synchronized than human-driven markets.
  - Example cited: the 2010 flash crash as an episode where automated systems affected markets before policymakers could intervene.
- Distributional risks:
  - AI can expand access (examples: Kenya’s M-Pesa transaction footprints used for AI-based scoring; India’s digital identity systems paired with machine learning for transfers and microloans).
  - AI can also entrench exclusion where data poverty or biased datasets leave women, rural populations, or informal workers invisible to algorithms.

### Resilience and accountability
- Self-preservation law translates into technical and institutional resilience:
  - Technical resilience: redundancy, monitoring, testing under extreme scenarios.
  - Institutional resilience: openness so regulators can audit AI decisions even when proprietary code is involved.
- Practical steps and capabilities:
  - Regulators need skills and tools to validate and challenge companies’ AI models.
  - Financial institutions should appoint AI risk officers analogous to chief risk or compliance officers to ensure explainability and auditability.
- Principle: “If an algorithm affects financial stability, it should be open to supervisory scrutiny.”
- Secrecy breeds fragility; complexity without transparency can lead to collapse (reference to the global financial crisis of 2008 as a reminder).

### Obedience and judgment: limits of automation
- AI optimizes given goals (profit, accuracy, compliance); blind obedience can produce morally problematic trade-offs:
  - A credit model minimizing default risk may exclude high-risk groups and deepen inequality.
  - A trading algorithm maximizing returns may exploit micro-price signals and destabilize markets.
- Policy implication: AI should remain a tool, not a decision-maker; human discretion and judgment remain essential in monetary policy, supervision, and crisis management.
- Models trained on past data can fail to predict new shocks; both machines and humans must learn together and be aware of mutual blind spots.

### A higher law: trust as the social contract of money
- Trust is the invisible infrastructure of finance; AI must strengthen rather than undermine it.
- When aligned with public interest, AI can:
  - Detect fraud faster.
  - Make supervision more proactive.
  - Extend financial access to previously excluded populations.
- When AI is biased, unstable, or unaccountable, it threatens the foundation of the financial system.

### Global coordination and institutional roles
- AI moves faster than regulation and crosses borders more quickly than money; cross-border coordination is essential to avoid fragmentation.
- Institutions and suggested roles:
  - Financial Stability Board (FSB), Bank for International Settlements (BIS), and the IMF: explore frameworks for responsible AI, emphasizing fairness, explainability, accountability, and proportionality.
  - BIS Innovation Hub: prototype tools to help supervisors analyze large datasets and detect anomalies.
  - IMF: use surveillance and technical assistance to help countries identify AI-related financial risks, share best practices, avoid a digital divide in supervision, and attract skilled professionals from research and fintech communities.
  - BIS: potential repository of supervisory algorithms to allow regulators to collaborate on open-source models.
  - World Bank and regional development institutions: build AI capacity and digital infrastructure in emerging markets; integrate ethical AI standards into digital finance ecosystems.
- Objective: digital multilateralism so AI benefits extend beyond advanced economies; no country can manage these dynamics alone.

### Policy recommendations and governance priorities
- Adopt an ethic for AI in finance based on the hierarchy adapted from Asimov:
  - Safety before obedience.
  - Transparency before secrecy.
  - Trust before profit.
- Concrete governance measures:
  - Test AI under extreme scenarios and enforce redundancy and continuous monitoring.
  - Ensure regulatory access to algorithms affecting financial stability, even when proprietary.
  - Appoint AI risk officers within financial institutions.
  - Build regulatory AI literacy and supervisory toolkits.
  - Promote international cooperation to develop a global set of principles for AI in finance, akin to the Basel Core Principles for banking.
  - Use IMF and World Bank technical assistance and financing instruments to reduce a digital divide in supervision and inclusion.

### Conclusion: maintaining human judgment
- The future of finance will increasingly be written in code, but principles behind it must remain human.
- Embrace intelligent governance rather than blind automation so AI becomes a stabilizing force that extends inclusion, enhances oversight, and strengthens monetary legitimacy.
- Closing thought: “The machines are learning. So must we.”

*Biagio Bossone — “Old Laws for New Machines” (F&D, March 2026).*

---


_Source: https://www.imf.org/-/media/files/publications/fandd/article/2026/03/bossone.pdf_