Industries / Finance and business services / Finance and banking
Credit unions
A credit union is a financial cooperative owned by its members: it takes their savings, lends them back to members, and is run by a board the members elect.
1850
Hermann Schulze-Delitzsch, a former district judge, set up a Vorschussverein, a loan society, in the Prussian town of Delitzsch, a year after a raw-materials association for the town's shoemakers; the Allgemeine Deutsche Biographie of 1891 calls it "der erste Vorschußverein", the first such society.
39
Milestones achieved across 7 domains.
9
To watch for: announced, not yet confirmed outside.
14
Open deliverables. Nobody has shipped them.
In one minute
Where it started
Credit unions began as self-help loan societies in mid-nineteenth-century Germany, Hermann Schulze-Delitzsch's in towns from 1850 and Friedrich Wilhelm Raiffeisen's in villages from 1864, and in both the members answered together for the society's debts. Alphonse Desjardins founded the first caisse populaire in Lévis, Quebec, in 1900, and a parish credit union in Manchester, New Hampshire, followed in 1908 or 1909. Massachusetts passed a credit union law in 1909, and the US Federal Credit Union Act of 1934 opened federal charters in every state.
Where it stands
The World Council of Credit Unions counted about 61,838 credit unions and financial cooperatives in 93 countries at the end of 2025, with nearly 420 million members and US$3.98 trillion in assets. In the United States, 4,214 federally insured credit unions served 146.1 million members in mid-2026, 156 fewer credit unions than a year earlier, as small ones merge into larger ones. The costliest lessons came from concentration and from insiders, and lately from the outside firms that run credit unions' computers: in the first year of NCUA's cyber incident rule, 7 in 10 reports involved a vendor.
What is open
Most open jobs are public tests and open tables built from records that regulators already publish but nobody has assembled: decades of quarterly reports, inspector general reviews of failures, merger notices, membership rules and the vendors each credit union depends on. Others are test sets of made-up members, books and conversations that show whether AI models and agents answer members correctly, catch insider theft and coached scam victims, or carry a conversion across without losing a balance. None needs real member data or real money.
Domains
AchievedWatch forOpen
Domain
Progress
Last achieved
Watch for
Open
Membership and accounts
A credit union belongs to the people who save and borrow there: each member buys a share to join, members elect the board, and their savings and checking balances are legally shares in the cooperative.
7 · 1 · 2
Last achievedOverdraft data withdrawn2025
Watch forAssociations that sell to members
2
openLending
Credit unions lend members' savings back to members, mostly for cars, homes and personal needs, and some lend to members' businesses.
5 · 1 · 2
Last achievedCredit builder loans tested2022
Watch forAI underwriting for small lenders
2
openCore systems
The core is the system of record that holds every member's shares and loans and posts each transaction; nearly every credit union now rents it, and the computers it runs on, from an outside firm.
5 · 1 · 2
Last achievedClients drift from the leader2025
Watch forOne cloud core, one region
2
openDigital banking
Members reach their accounts without a branch: through websites, phone apps and text menus, by sending money that arrives in seconds, and by letting other apps read their account data.
4 · 2 · 2
Last achievedKenyan SACCO members go mobile2024
Watch forOpen banking rule rewritten
2
openFraud
Theft from credit unions and from their members, by insiders, outside criminals and scammers, and the checks and shared warnings that catch it.
5 · 1 · 2
Last achievedCyber incidents within 72 hours2023
Watch forName checks before paying
2
openRegulation and capital
The laws, supervisors and insurance funds that protect members' savings, and the capital, meaning a credit union's own reserves, that it must hold against losses.
7 · 1 · 2
Last achievedCapital weighted by risk2022
Watch forRules cut in rounds
2
openShared services
Most credit unions are too small to do everything alone, so they pool work: emergency cash and payments through credit unions for credit unions, branches that serve each other's members, and service companies they own together.
6 · 2 · 2
Last achievedUK allows shared service firms2026
Watch forA treasury pool in Ireland
2
open
Advances to build with
3 research notes

Mistral Large 4
A trillion open parameters
Mistral published its Large 4 AI model, a trillion numbers by its count, free to download and sell, though firms that rent out models need a separate license.

Muse Ember
The jobs agents repeat
Meta's Muse Ember matched Claude Haiku 5.5 on an outside firm's tests at a third of the cost per task, though both list the same prices for text in and out.

Microsoft ThinkingBox
All twenty times
By Microsoft's own count, 50 steps of training took an AI model from passing 38 of 507 test tasks on all 20 tries to passing 107.
Open in credit unions
3 of 14 open deliverables
Open · Membership and accounts
Merger notices as open data
An open table of every federally insured US credit union merger put to a member vote since 2018, taken from the member notices that NCUA posts in public dockets. For each merger: the two credit unions, their assets and members, the stated reason, the vote date, any special dividend to members, and each disclosed merger-related payment or raise for named executives and directors, with the page of the notice for every figure. A second pass checks each notice for the items the rule requires and lists what is missing.
Built with what exists
The regulations.gov API, version 4, which returns each merger docket's documents, for example the "Member Notice" in docket NCUA-2026-2345, posted on 17 November 2026; about 1,000 NCUA dockets match "merger" in the API; pdftotext or pdfplumber; a low-cost AI model run as an extraction agent (see the Note "The jobs agents repeat"); a second model that reads each notice against the rule's list of required items (see the Note "Seven more flagged").
Open · Lending
AI lending claims, checked
An outside test of whether credit unions that adopted AI loan underwriting changed their lending results. Take credit unions whose adoption date is public, from vendor case studies and press releases, and compare their auto and personal loan growth, delinquency (loans 30 or more days late) and net charge-offs (loans written off, less recoveries) quarter by quarter before and after adoption, against similar credit unions that announced no adoption.
Built with what exists
NCUA's quarterly call report files, which give each credit union's loan balances, delinquencies and charge-offs by loan type; adoption dates from Zest AI's and other vendors' published case studies; Python with pandas, and the R package did for staggered difference-in-differences.
Open · Core systems
Map the vendors behind vendors
An open table, one row per US credit union, of the outside firms it depends on in four layers: its core, its online and mobile banking, its card processing and the firm that hosts its systems. Each cell carries the public evidence and the date it was seen: a vendor's press release or client list, the web address of the login page, the developer named in the app store, a job posting or an annual report. From the table, compute how many credit unions and members would lose service if each firm, including a firm one step removed, went down.
Built with what exists
NCUA's quarterly call report files for the list of credit unions with their assets and members; a browser agent built on the open-source Browser Use library to visit each credit union's website and record the login page's address and the app's developer (see the Note "Deleting 80% to start over"); a language model to read press releases and job postings into fields, each with a quoted excerpt (see the Note "The jobs agents repeat"); SQLite or a spreadsheet.
Where it sits
Broader