Propositions · 12 propositions · 173 records · 1997–2026

AI and money: what central banks and financial authorities have said, 1997–2026

Which located texts by central banks, the BIS, the FSB, the IMF, standard setters and supervisors, from 1997 to 2026, state, contest or bear on propositions about machine intelligence in financial stability, monetary policy, payments and supervision? For each proposition, which text is the earliest located, whoever speaks in it, and which is the earliest located text in an institution’s own voice? Which institutional events do these bodies state?

This collection records what central banks, the Bank for International Settlements, the Financial Stability Board, the IMF, standard setters and supervisors have said about machine intelligence. Twelve propositions cover financial stability, monetary policy, payments, and supervision and central banks’ own work. Each text is labelled by whom it speaks for. An institution’s own text is the institution’s. A speech or remarks are the speaker’s, with the speaker’s office and any statement that the views are the speaker’s own. A staff paper is its authors’, and a committee report the committee’s. None of them is the institution’s policy unless the institution’s own text says so. One earliest located text is named for each proposition, whoever speaks in it. Where that text is a speech, a staff paper or a committee report, the earliest located text in an institution’s own voice is shown beside it. A list of institutional events follows the propositions: plans announced, work stated as under way and work stated as done, each dated by the text that states it. Texts are listed in date order; the order records no influence between them.

Download research starter

The original sources first, in date order, then every proposition on this page with its criterion, earliest located text and records. Each record gives its original source and the place in it, its role, register and dates, the qualification and the missing element of a context record, then the Selekoa identifiers. Stored paraphrases and permitted short quotes are included. Original documents are not.

Original sources for a reference manager. Every source this export cites, for Zotero, EndNote and similar tools. Cite the original documents, not Selekoa, and check each entry against them.

Download CSV

Propositions by author →

How we searched

The collection records located texts dated from 1997 to 2026 by the bodies named below. “Earliest located” means the earliest text found by these searches that meets the criterion, judged by the date of its wording among texts of every voice. It is not a claim that no earlier text exists. A report of what others said, such as a summary of a workshop or of respondents to a consultation, never holds a headline. The systematic searches cover 2016 to 2026; a bounded search of the same bodies before 2016 followed. The headlines are only as good as that start date and that list of bodies, and they are weakest where the located date is close to 2016.

Venues and years searched

2016 to 2026, systematic

The BIS: Annual Economic Reports, working papers, FSI Insights, Irving Fisher Committee bulletins, Innovation Hub project pages and speeches by BIS management. The Basel Committee, the Financial Stability Board, the IMF, the ESRB, the EBA and EIOPA, where they state a proposition.
The ECB and ECB Banking Supervision, the Bank of England with the PRA, the Bank of Finland, the Deutsche Bundesbank, the Riksbank, De Nederlandsche Bank, Norges Bank and Danmarks Nationalbank.
The Federal Reserve Board, the People’s Bank of China, the Bank of Japan, the Monetary Authority of Singapore, the Hong Kong Monetary Authority and the Reserve Bank of India.
Each body’s own publication search or speech index, for “artificial intelligence”, “machine learning”, “AI”, “generative AI”, “AI agents” and “agentic”, and the texts these cite.

Before 2016, bounded

Searches restricted to the same bodies’ sites, and unrestricted searches with the bodies’ names, for each proposition’s terms.
Neural-network and machine-learning papers in the ECB, Bank of England, Bundesbank, Federal Reserve and Bank of Finland series; AI in BIS and Basel Committee papers on supervision; speeches of 2014 and 2015 on AI and monetary policy or financial stability.

Methods

Sources were retrieved directly by script, one request at a time, from 2026-11-02 to 2026-11-04, and again at admission on 2026-11-06. Each retrieved text is identified by its SHA-256 hash in the source record.
Every quotation was matched mechanically against the retrieved text, both times. Paraphrases are labelled as such. Quotations in Chinese, Danish, Finnish, Norwegian and Swedish keep the original with a working English translation.
A fresh-context audit by a Claude model re-read a sample of records. Two checks by an OpenAI model followed, each answering blind before it compared its answers with the records. The maintainer decided the disagreements on 2026-11-03 and 2026-11-04.

Known gaps

Securities regulators enter only through IOSCO and the bodies named. Two speeches by Scott W. Bauguess of the US Securities and Exchange Commission (21 October 2016 and 21 June 2017), on the SEC’s use of machine learning in supervision, are not recorded.
Central banks outside the named list are not recorded: a Czech National Bank working paper of December 2005 and a Bank of Canada working paper of January 1999, both on neural-network forecasting, are known and not recorded.
BaFin’s report “Big data meets artificial intelligence” (July 2018) is not recorded, since BaFin alone is not among the bodies named; its joint paper of 2021 with the Bundesbank is recorded.
The Bank of England’s “One Bank Research Agenda” (February 2015), whose Theme 4 describes machine-learning analysis of text, is recorded on J05 from 2026-11-07, as its earliest located institutional text. It was read by a model of another family in an archive capture of March 2015, the only copy found; its record paraphrases that reading. The Bundesbank’s discussion paper of November 2020 was not read.
Not read: an IMF speech of 6 June 2024 and two IOSCO reports (the sites refused retrieval), two speeches by Fan Yifei of the People’s Bank of China (2020, 2021) and the record page of the Bank’s standard for evaluating AI algorithms.
Industry papers convened by the Monetary Authority of Singapore (MindForge 2024, SAFR 2026), which the authority labels industry-led, are not recorded.
The archives were not searched year by year in full text. Press reports, interviews, video without a transcript and social media were used for discovery only. A search that finds nothing does not show that nothing exists.

A text found later that is earlier than an earliest located text becomes the new earliest located text, and the earlier headline stays on record as a correction.

How to read a record

Articulates: A passage meets the proposition’s criterion.
Contests: A passage disputes the proposition or a premise it depends on.
Bears on: A passage comes close without meeting the criterion; the record names what is missing.
Earliest located: The earliest articulation the searches found, dated by its wording. It is not a claim that no earlier text exists. The earliest located text that contests a proposition is chosen by the same rule.
Earliest located use of the name: The earliest located text that uses the proposition’s usual name. A title can carry the name without stating the proposition.
Engages a premise, states the contrary: A contesting passage engages a premise when its paragraph, message or dialogue turn gives a reason against it. Otherwise it states the contrary conclusion only.
Later restatement: The author restates a point made in an earlier located text of their own.
Dates: The first date is the one the work is dated by. “Wording attested” gives the date from which the quoted wording is known, when it survives only in a later copy. “Claimed” gives an earlier talk, draft or presentation that does not date the wording.
Voice: who a text speaks for: An institution’s own text is the institution’s. A speech or remarks are the speaker’s, attributed to the speaker and the office, with any statement that the views are the speaker’s own. A staff paper is its authors’. A committee report is the committee’s. Where a speech, a staff paper or a committee report is earliest located, the earliest located institutional text is shown beside it.
Register: the kind of text: Fiction, satire, a dream frame or a relay of another author is labelled. A passage in fiction records what the work says, not its author’s view. Unlabelled records are nonfiction.

Limits of this page

“Earliest located” means the earliest text found by the searches described under “How we searched”. It is not a claim that no earlier text exists.
A record documents what a text says, in its voice. It is not evidence that a risk has materialised, that a policy was adopted, or that the proposition is true.
A speech, a staff paper or a committee report is not the institution’s position unless the institution’s own text says so.
An event records what a text states about an authority’s own work. Announced stays announced, and work stated as under way is not shown as done.
A machine selected and classified these records, and machine checks from two model families re-examined them.

JA · Machine intelligence and financial stability

Whether the use of AI by financial firms could make their behaviour more alike, concentrate their dependence on a few providers, add cyber, fraud and operational risk, or make their models harder to explain and supervise.

J01 AI raises financial-stability risks through correlated behaviour

Criterion. Widespread use of similar AI or machine-learning models, data or techniques by financial firms could make their decisions more alike (herding, uniformity, correlated trading or lending), which could amplify shocks, procyclicality or market volatility.

Earliest located: Haruhiko Kuroda · AI and the Frontiers of Finance · 2017-04-13 · Speech or remarks
Earliest located institutional text: Financial Stability Board · Artificial intelligence and machine learning in financial services: Market developments and financial stability implications · 2017-11-01 · Institution’s own text
Earliest located text that contests it: Board of Governors of the Federal Reserve System · Financial Stability Report, November 2025 (Box 1.1, ‘Artificial Intelligence and Algorithmic Trading’) · 2025-11 · Institution’s own text

Filled dots articulate, open dots contest, small dots bear on the proposition; a large dot is an earliest located text. A line below a dot runs to the date from which its wording is attested.

Texts that articulate or contest it (17)

Texts that bear on it without meeting the criterion (5)

J02 Reliance on a few AI providers creates concentration risk

Criterion. Dependence of many financial firms on a small number of providers of AI models, data, cloud computing or related services creates concentration or third-party risk for the financial system, such that the failure or disruption of one provider could affect many firms at once.

Earliest located: Financial Stability Board · Artificial intelligence and machine learning in financial services: Market developments and financial stability implications · 2017-11-01 · Institution’s own text

Filled dots articulate, open dots contest, small dots bear on the proposition; a large dot is an earliest located text. A line below a dot runs to the date from which its wording is attested.

Texts that articulate or contest it (13)

Texts that bear on it without meeting the criterion (3)

J03 AI increases cyber, fraud and operational risks in finance

Criterion. AI, including generative AI, increases the capabilities available for cyberattacks, fraud, impersonation (such as deepfakes) or market manipulation against financial firms, customers or markets, or adds operational risks that financial authorities should take into account.

Earliest located: Financial Stability Board · Artificial intelligence and machine learning in financial services: Market developments and financial stability implications · 2017-11-01 · Institution’s own text

Filled dots articulate, open dots contest, small dots bear on the proposition; a large dot is an earliest located text. A line below a dot runs to the date from which its wording is attested.

Texts that articulate or contest it (15)

J04 The opacity of AI models complicates risk management and supervision

Criterion. The limited explainability or interpretability of AI or machine-learning models (the ‘black box’ problem) makes it harder for financial firms to manage their risks, or for supervisors to assess them or hold firms accountable.

Earliest located: Financial Stability Board · Financial Stability Implications from FinTech: Supervisory and Regulatory Issues that Merit Authorities’ Attention · 2017-06-27 · Institution’s own text

Filled dots articulate, open dots contest, small dots bear on the proposition; a large dot is an earliest located text. A line below a dot runs to the date from which its wording is attested.

Texts that articulate or contest it (14)

Texts that bear on it without meeting the criterion (1)

Bears on · Institution’s own textBank for International Settlements · III. Big tech in finance: opportunities and risksThe BIS Annual Economic Report 2019 (chapter III) states in an endnote that complex and opaque algorithms render biases in machine-learning credit scoring particularly difficult to detect and therefore to prevent.Missing: Does not say whose task (firms’ risk management or supervisors’ assessment) the opacity makes harder; the passage concerns detecting discriminatory bias in big tech algorithms.2019-06-30 Inspect claim & evidence ↗Document page →

JB · Monetary policy and the economy

Whether AI can improve forecasting and analysis at central banks, and whether AI’s effects on the economy matter for monetary policy.

J05 AI can improve nowcasting and policy analysis

Criterion. Machine-learning or AI techniques can improve nowcasting, forecasting, the processing of large or unstructured data, or other economic and policy analysis used by central banks.

Earliest located: Paul McNelis; Peter McAdam · Forecasting inflation with thick models and neural networks · 2004-04 · Staff paper
Earliest located institutional text: Bank of England · One Bank Research Agenda (discussion paper) · 2015-02-25 · wording attested 2015-03-19 · Institution’s own text

Filled dots articulate, open dots contest, small dots bear on the proposition; a large dot is an earliest located text. A line below a dot runs to the date from which its wording is attested.

Texts that articulate or contest it (14)

Texts that bear on it without meeting the criterion (1)

Bears on · Speech or remarksLisa D. Cook · Generative AI, Productivity, the Labor Market, and Choice BehaviorIn a speech of 22 September 2023, Lisa D. Cook, Governor of the Federal Reserve Board, said that in the policy arena an AI black box with no insight into the decision-making process is of limited value, and that she looks on model-generated forecasts with a skeptical eye if they are not coupled with a plausible explanation of their driving factors.Missing: A statement that AI or machine learning does not, or cannot, improve policy analysis; the passage conditions the value of AI forecasts on an explanation of their drivers.2023-09-22 Inspect claim & evidence ↗Document page →

Headlines replaced

Chakraborty and Joseph, Bank of England Staff Working Paper No. 674 (September 2017), headlined as earliest located by the research pack of 2026-11-02. Earlier text found before admission, in the bounded search before 2016: McNelis and McAdam, ECB Working Paper No. 352 (April 2004). Recorded 2026-11-06.

J06 AI’s economic effects matter for monetary policy

Criterion. The effects of AI on productivity, labour markets, prices, investment or the neutral rate of interest are relevant to the conduct of monetary policy, so that central banks need to assess or respond to them.

Earliest located: Andrew G Haldane · Labour’s Share · 2015-11-12 · Speech or remarks
Earliest located institutional text: Bank for International Settlements · III. Artificial intelligence and the economy: implications for central banks · 2024-06-30 · Institution’s own text

Filled dots articulate, open dots contest, small dots bear on the proposition; a large dot is an earliest located text. A line below a dot runs to the date from which its wording is attested.

Texts that articulate or contest it (15)

Headlines replaced

Carney, Whitaker Lecture (14 September 2018), headlined as earliest located by the research pack of 2026-11-02. Earlier text found before admission by a check of another model family: Haldane, “Labour’s Share”, a speech of 12 November 2015. Recorded 2026-11-06.

JC · Payments and money

Whether AI agents will change how payments are made, and whether their transactions bear on tokenised money and stablecoins. No located text meets the second criterion; the texts that come close are shown as context.

J07 AI agents will change payment flows

Criterion. AI agents acting for people or firms will or could initiate, negotiate or execute payments and transactions, changing how payments are made or the demands placed on payment systems.

Usual names: “agentic payments”, “agentic commerce”

Earliest located: Iñaki Aldasoro; Ajit Desai · AI agents for cash management in payment systems · 2025-11-10 · Staff paper
Earliest located institutional text: Monetary Authority of Singapore · MAS Partners Industry to develop Safeguards for AI Agents in Finance · 2026-07-03 · Institution’s own text
Earliest located use of the name: Sarah Breeden · Agents of change · 2026-06-30 · Speech or remarks · “agentic commerce”

Filled dots articulate, open dots contest, small dots bear on the proposition; a large dot is an earliest located text. A line below a dot runs to the date from which its wording is attested.

Texts that articulate or contest it (7)

Texts that bear on it without meeting the criterion (3)

Headlines replaced

Waller, a speech at Sibos (29 September 2025), headlined as earliest located by the research pack of 2026-11-02. On a check of another model family, the speech names AI agents executing tasks, not payments or transactions, and is recorded as context; a report of workshop participants’ ideas cannot hold the headline. BIS Working Paper No. 1310 (10 November 2025) is earliest located. Recorded 2026-11-06.

J08 AI agents’ transactions bear on tokenised money and stablecoins

Criterion. Transactions initiated by AI agents would use, require or favour programmable or tokenised forms of money, stablecoins or tokenised deposits, or AI would interact with such forms of money in ways that matter for the monetary system.

Earliest located: No located text meets the criterion.

Filled dots articulate, open dots contest, small dots bear on the proposition; a large dot is an earliest located text. A line below a dot runs to the date from which its wording is attested.

Texts that articulate or contest it (0)

Texts that bear on it without meeting the criterion (4)

JD · Supervision, risk management and central banks’ own work

Whether supervisors should use AI, whether AI improves firms’ risk management, whether central banks need their own AI capacity, and whether existing financial rules broadly cover AI.

J09 Supervisors should use AI

Criterion. Financial supervisors or regulators can, or should, use AI or machine-learning tools (supervisory technology, ‘suptech’) to carry out supervision, including detecting misconduct, money laundering or emerging risks.

Usual name: “suptech”

Earliest located: Ranjana Sahajwala; Paul Van den Bergh · Supervisory risk assessment and early warning systems · 2000-12 · Staff paper
Earliest located institutional text: Financial Stability Board · Financial Stability Implications from FinTech: Supervisory and Regulatory Issues that Merit Authorities’ Attention · 2017-06-27 · Institution’s own text · “suptech”
Earliest located use of the name: Financial Stability Board · Financial Stability Implications from FinTech: Supervisory and Regulatory Issues that Merit Authorities’ Attention · 2017-06-27 · Institution’s own text · “suptech”

Filled dots articulate, open dots contest, small dots bear on the proposition; a large dot is an earliest located text. A line below a dot runs to the date from which its wording is attested.

Texts that articulate or contest it (13)

Headlines replaced

The FSB report on the financial stability implications of FinTech (27 June 2017), headlined as earliest located by the research pack of 2026-11-02. Earlier text found before admission, in the bounded search before 2016: Sahajwala and Van den Bergh, Basel Committee Working Paper No. 4 (December 2000), whose “artificial intelligence” is expert-system software. The FSB report stays the earliest located institutional text and the earliest located use of “SupTech”. Recorded 2026-11-06.

J10 AI can improve financial firms’ risk management

Criterion. AI or machine-learning models can improve financial firms’ measurement or management of risk, such as credit assessment, fraud detection, anti-money-laundering controls or market-risk modelling.

Earliest located: Susan M. Phillips · Risk management · 1997-11-04 · Speech or remarks
Earliest located institutional text: Financial Stability Board · Financial Stability Implications from FinTech: Supervisory and Regulatory Issues that Merit Authorities’ Attention · 2017-06-27 · Institution’s own text

Filled dots articulate, open dots contest, small dots bear on the proposition; a large dot is an earliest located text. A line below a dot runs to the date from which its wording is attested.

Texts that articulate or contest it (12)

Texts that bear on it without meeting the criterion (1)

Bears on · Institution’s own textBasel Committee on Banking Supervision · Sound Practices: Implications of fintech developments for banks and bank supervisorsThe Basel Committee’s ‘Sound Practices’ report of February 2018 states that the accuracy and validity of AI/ML models increasingly used for the termination of credit limits is as yet unproven.Missing: A statement that AI or machine-learning models do not, or cannot, improve risk management; the passage calls their accuracy and validity for the termination of credit limits unproven.2018-02 Inspect claim & evidence ↗Document page →

J11 Central banks need to build their own AI capacity

Criterion. Central banks need to invest in their own AI capabilities (data, computing, skills or staff), or to cooperate or pool resources with other central banks, to make use of AI or keep pace with its adoption.

Earliest located: Sebastian Doerr; Leonardo Gambacorta; Jose Maria Serena · Big data and machine learning in central banking · 2021-03 · Staff paper
Earliest located institutional text: Bank for International Settlements · III. Artificial intelligence and the economy: implications for central banks · 2024-06-30 · Institution’s own text

Filled dots articulate, open dots contest, small dots bear on the proposition; a large dot is an earliest located text. A line below a dot runs to the date from which its wording is attested.

Texts that articulate or contest it (8)

Texts that bear on it without meeting the criterion (1)

Bears on · Staff paperEl Bachir Boukherouaa; Ghiath Shabsigh; Khaled AlAjmi; Jose Deodoro; Aquiles Farias; Ebru S. Iskender; Alin T. Mirestean; Rangachary Ravikumar · Powering the Digital Economy: Opportunities and Risks of Artificial Intelligence in FinanceIn the IMF Departmental Paper of 22 October 2021, Boukherouaa, AlAjmi, Deodoro, Farias and Ravikumar write that central banks have been slow to adopt AI/ML, attributing this in part to a lack of adequate capacity, and that risks would be amplified if a central bank lacks the resources and skills to operate AI/ML safely.Missing: An explicit statement that central banks need to invest in, or pool, AI capabilities; the conclusion’s call for capacity building is addressed to regulators generally.2021-10-22 Inspect claim & evidence ↗Document page →

J12 Existing financial rules broadly cover AI

Criterion. Existing financial regulatory and supervisory frameworks (for example technology-neutral rules on model risk, governance, outsourcing or consumer protection) broadly apply to the use of AI by financial firms, so that AI does not, or does not yet, require a separate regulatory framework.

Earliest located: Lael Brainard · What Are We Learning about Artificial Intelligence in Financial Services? · 2018-11-13 · Speech or remarks
Earliest located institutional text: De Nederlandsche Bank · General principles for the use of Artificial Intelligence in the financial sector · 2019-07-25 · Institution’s own text
Earliest located text that contests it: 中国人民银行 (People’s Bank of China) · 金融科技（FinTech）发展规划（2019—2021年） · 2019-08 · Institution’s own text

Filled dots articulate, open dots contest, small dots bear on the proposition; a large dot is an earliest located text. A line below a dot runs to the date from which its wording is attested.

Texts that articulate or contest it (17)

Texts that bear on it without meeting the criterion (9)

Institutional events

What an authority’s own text, or one of its officials, states about its work on AI: a plan announced, work stated as under way, or work stated as done. Each is dated by the text that states it. A statement is not evidence that the work was carried out or what it achieved.