Industries / Finance and business services / Finance and banking
Compliance and reporting
This work keeps the books, files the reports that regulators and investors rely on, and shows that a bank holds enough capital and follows the rules.
7
Milestones achieved, from 1494 to 2023.
2
To watch for: announced, not yet confirmed outside.
2
Open deliverables. Nobody has shipped them.
The road
AchievedWatch forOpen
Achieved
1494
Double-entry bookkeeping in print
In double-entry bookkeeping every transaction is written down twice, as a debit in one account and a credit in another, so a clerk can check the books by seeing whether they balance. Luca Pacioli went to Venice in 1494 to publish his Summa, a mathematics book that the historian S. A. Jayawardene describes as including "a treatise on double-entry bookkeeping", by the account of the University of St Andrews' MacTutor history. Pacioli wrote the method down; the source does not say he invented it.
Credit
Luca Pacioli
1934
Companies must tell investors
The US Securities Act of 1933 requires that investors get key facts about securities offered for sale to the public and bans fraud in their sale. The Securities Exchange Act of 1934 created the Securities and Exchange Commission (SEC) and let it require regular reports from companies whose shares trade publicly. The SEC says it was founded "at the height of the Great Depression".
Credit
US Congress; the SEC
1988
One capital rule for banks
Central bank governors of the Group of Ten countries set up what became the Basel Committee on Banking Supervision at the end of 1974, after disturbances in currency and banking markets that included the failure of Bankhaus Herstatt in West Germany. In July 1988 the committee released the Basel Capital Accord, which required capital of at least 8% of risk-weighted assets, meaning loans and other holdings counted according to how risky each kind is, by the end of 1992. The Bank for International Settlements (BIS), which hosts the committee, says it was adopted in "virtually all countries with active international banks".
LessonThe 2007–09 crisis exposed too much borrowing and too little cash at banks, so Basel III (December 2010) added a leverage ratio and two liquidity rules.
Credit
Basel Committee on Banking Supervision
2002
Auditors get an overseer
After the collapses of Enron, WorldCom and the accounting firm Arthur Andersen, the Sarbanes-Oxley Act became law on 30 July 2002. It set up the Public Company Accounting Oversight Board (PCAOB) "to oversee the audit of public companies", and requires chief executives and finance chiefs to certify their companies' reports and management to assess its internal controls, the checks that keep the books accurate. The SEC Historical Society notes that before the act "the audit profession had been self-regulated". In July 2004 the SEC charged Enron's former chairman, Kenneth Lay, with fraud and insider trading.
LessonCongress moved oversight of auditors from their own profession to an outside board.
Credit
US Congress; the PCAOB
2009
Financial reports become data
The SEC made filing on its EDGAR system mandatory in phases from 1993 to 6 May 1996, when the last group of US companies moved online. On 18 December 2008 it voted to require interactive data, tags a computer can read on each figure in a report, in the format called XBRL. About 500 of the largest companies started with periods ending on or after 15 June 2009. The EU followed: annual reports of companies listed on its regulated markets must be in XHTML with Inline XBRL tags for financial years beginning on or after 1 January 2020.
Credit
SEC; European Commission and the European Securities and Markets Authority (ESMA)
2012
One code for each firm
The Legal Entity Identifier (LEI) is a 20-character code, under the standard ISO 17442, that names each company taking part in financial deals. The G20 endorsed the LEI Charter in 2012, and the Financial Stability Board set up the Global Legal Entity Identifier Foundation (GLEIF) in 2014. GLEIF's timeline says the idea got its go-ahead in 2008 from Jean-Claude Trichet, then president of the European Central Bank, "under the impression of his Lehman experiences", and that its public index passed 1 million records in 2017 and 2 million in 2022.
Credit
G20 and the Financial Stability Board; GLEIF
2023
Risk data still falls short
In 2013 the Basel Committee issued principles, known as BCBS 239, for how banks gather data on their risks and report it to their boards. In November 2023 it reported on 31 global systemically important banks, the largest banks whose failure would harm the world system, nearly ten years after the principles and seven years after the date banks were expected to comply. Deloitte's reading of the report is that "only two are fully compliant", up from none in 2019, and it cites "fragmented IT landscapes, legacy systems, and manual processes". The committee wrote that "additional work is required at all banks".
LessonWriting a data rule is fast, while rebuilding the old systems and manual steps behind bank reports takes more than a decade.
Credit
Basel Committee on Banking Supervision; summary by Deloitte
Watch for
Audits run with AI agents
EY says it has built a framework of AI agents into EY Canvas, its global audit platform, which 130,000 of its assurance staff use on 160,000 audit engagements in more than 150 countries and territories, and which processes more than 1.4 trillion lines of journal-entry data a year. EY expects the framework "to support all end-to-end audit activities by 2028". The release gives no measured results.
Achieved whenAn audit regulator's inspection reports, such as the PCAOB's, or an independent study publish error or deficiency rates for audits done with the agents next to audits done without them.
Claimed by
EY, April 2026
One statistics return, euro area
The European Central Bank (ECB) plans the Integrated Reporting Framework (IReF), "a single standardised reporting framework that is directly applicable to banks resident in the euro area", for the statistics banks send their central banks. A one-year pilot is to start in the second quarter of 2030 and first official reporting in the second quarter of 2031, with a year of reporting in both the old and new ways. It is to run in a European cloud, and the timeline depends on a public consultation in the second half of 2027.
Achieved whenEuro-area banks file under IReF and the older statistical reports it covers are switched off after the parallel year.
Claimed by
ECB, June 2026
Open
01
Predict the SEC's comment letters
A public benchmark that pairs company filings, such as annual reports on Form 10-K, with the comment letters SEC staff later sent about them. A model reads only the filing as first filed and lists the issues it would raise. Scoring counts how many of the staff's comments it matched, and how many of its points the staff never raised. All the data is public on EDGAR.
The briefCredit · unclaimed02
Complaints that warn regulators
A public test of whether consumer complaints warned of problems before a regulator acted. Take the US Consumer Financial Protection Bureau's (CFPB) public complaint database, freeze copies as they stood before each of a set of past CFPB enforcement actions, and add matched company-and-product pairs that never led to an action. Score any method on how many actions it flags, how many months early, and how many false alarms it raises. Use public records only, and report results as rates, not as lists of flagged companies.
The briefCredit · unclaimed
Open 01 · the brief
Predict the SEC's comment letters
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
A public benchmark that pairs company filings, such as annual reports on Form 10-K, with the comment letters SEC staff later sent about them. A model reads only the filing as first filed and lists the issues it would raise. Scoring counts how many of the staff's comments it matched, and how many of its points the staff never raised. All the data is public on EDGAR.
Why it is still open
SEC staff review every reporting company at least once every three years and publish their letters and the companies' replies no sooner than 20 business days after a review ends, so years of matched pairs exist. The closest study found, by James Ryans of London Business School (Review of Accounting Studies, 2021), sorted letters after the fact to predict restatements and write-downs; it does not predict staff comments from the filing. Searches found no published benchmark for this.
Built with what exists
EDGAR's free full-text search and data downloads; an AI model that reads long documents, such as Claude Fable 5.2 (see the Note "Three working devices") or the open-weight Mistral Large 4 (see the Note "A trillion open parameters"); a reviewer that cites the rule behind each point it raises (see the Note "Findings that cite notes").
First step
Download 20 recent letter sets from reviews of 10-K filings, split each letter into numbered comments, and record the filing and page each comment refers to.
Done looks like
At least 200 filings from at least three years, each paired with its released letters and matched comment by comment by two people who agree on at least 90% of a 50-pair sample; published scoring code; and a results table giving, for at least two models and one keyword baseline, the share of staff comments matched and the number of unmatched model points per filing.
Open 02 · the brief
Complaints that warn regulators
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
A public test of whether consumer complaints warned of problems before a regulator acted. Take the US Consumer Financial Protection Bureau's (CFPB) public complaint database, freeze copies as they stood before each of a set of past CFPB enforcement actions, and add matched company-and-product pairs that never led to an action. Score any method on how many actions it flags, how many months early, and how many false alarms it raises. Use public records only, and report results as rates, not as lists of flagged companies.
Why it is still open
The CFPB says complaints help it "better regulate consumer financial products and services, enforce those laws judiciously", and its database still updates daily. The published model found, by Dhwani Vaishnav and colleagues (2024), predicts how fast and how companies respond to complaints, not later enforcement, and searches found no shared benchmark that links complaints to enforcement actions.
Built with what exists
The CFPB complaint database and its list of enforcement actions, which showed 386 actions in November 2026; Python; an AI model that returns probabilities for fixed-answer questions, to sort complaint text into issues (see the Note "A thousand questions, one pass"); a small model trained on one graphics chip (see the Note "One GPU to fine-tune").
First step
Pick ten enforcement actions filed between 2016 and 2020, download the complaint file, and build each snapshot by keeping only complaints the CFPB received before the action's filing date.
Done looks like
At least 50 past enforcement actions and 50 matched controls, each with a frozen complaint snapshot; published scoring code; and a results table for at least two methods, one of them a simple complaint-count baseline, giving actions caught, median months of warning, and false alarms per action caught.