How many AI incidents have banks publicly had since 2022, and what kind?
As of Oct 21, 2026, BankingNewsAI's register holds 62 publicly reported AI incidents at banks and fintechs dated from January 2022: 9 where the institution's own AI failed, was withdrawn or was reversed; 29 where criminals or others used AI against a bank or its customers; and 24 governance findings (bias, litigation, regulatory findings, AI-washing). The most common category is brand impersonation (11). By year: 2022, 10; 2023, 10; 2024, 11; 2025, 16; 2026, 15. 30 of the 62 records involve systems the institution or a regulator described as AI; 11 are automated systems not described as AI, and 21 are machine-learning systems not labelled AI on the record. The most recent entry is "U.S. Bank customers sue Pegasystems alleging its Voice AI secretly analysed their customer-service calls" (Oct 2, 2026). Every entry cites at least two independent publishers or one official or court source, and records the institution's own response. The register is new and incomplete: a bank without an entry has either been swept with nothing found (the date is stated on its page) or not been swept yet.
Data as of Oct 21, 2026 · computed from the register, not hand-counted · Markdown
When did AI incidents at banks happen, and whose AI was it?
How can the register be filtered?
By role
Institution's own AI (9)
AI used against the institution (29)
Governance finding (24)
By year
2022 (10)
2023 (10)
2024 (11)
2025 (16)
2026 (15)
By country
United States (27)
United Kingdom (8)
Several countries (3)
Australia (2)
India (2)
Italy (2)
South Korea (2)
Netherlands (2)
Sweden (2)
Brazil (1)
Canada (1)
Germany (1)
Dominican Republic (1)
Spain (1)
Hungary (1)
Norway (1)
New Zealand (1)
Poland (1)
Romania (1)
Singapore (1)
Thailand (1)
By status
Reported (11)
Confirmed by the institution (18)
Regulator finding (8)
Court filing (10)
Settled (8)
Resolved (4)
Disputed (3)
By category · Chatbot failure (1) · Credit decisioning (1) · Fraud model (1) · Trading automation (1) · Agent / automation (0) · AI outage (0) · AI job-cut reversal (2) · Withdrawn deployment (0) · Deepfake fraud (7) · Voice clone (8) · AI phishing (0) · Synthetic identity (1) · Account takeover (0) · Brand impersonation (11) · Prompt injection (1) · Data leak to an LLM (1) · Vendor AI breach (1) · Bias finding (3) · Litigation (11) · Regulatory finding (6) · AI-washing (2) · Other (4)
Every AI incident on record, newest first
How have banks responded to AI risk without an incident?
Not counted as incidents: dated precautions such as bans on public chatbots, paused deployments, regulator warnings and internal probes where no harm was reported. 40 items so far.
- RBC weighs blocking Meta's Muse AI agent; BMO CEO warns against giving agents banking credentials — other, Oct 20, 2026
- Bank of America adds 1,000 contractors to patch software flaws exposed by Anthropic's Mythos — other, Oct 14, 2026
- JPMorgan, Citi, Bank of America, Schwab and Fifth Third block or limit outside AI agents in customer accounts — other, Oct 13, 2026
- ASIC to review Australian banks' AI use cases and their impact on customers — regulator warning, Sep 30, 2026
- Six banks including Bank of America and NatWest warn AI shopping agents raise fraud and privacy risks — other, Sep 22, 2026
How is the register built, and what has been searched?
An incident is an event dated 2022 or later in which an AI system caused or contributed to harm or a reversal at a bank or fintech: its own AI failing or being withdrawn, AI used against it or its customers, or a governance finding. Each record needs two sources from two different publishers, or one official or court source; aggregator sites are not accepted; the institution's response is always recorded. Severity: 1 contained, no customer harm; 2 limited harm or a policy reversal; 3 material harm or a regulatory finding; 4 large losses, many customers, or a consent order; 5 systemic, criminal, or across several institutions.
Sweeps search news, regulator, court and security sources for one institution at a time. So far: 190 institutions swept between Oct 21, 2026 and Oct 21, 2026, 3793 queries, 111 candidate stories reviewed, 37 accepted as records. "No incident found" is always a dated statement: it says what a sweep on that date found in public reporting, not that nothing happened.
| Institution | Swept | Queries | Candidates | Accepted |
|---|---|---|---|---|
| Wintrust Financial Corporation | Oct 21, 2026 | 17 | 0 | 0 |
| Western Alliance Bancorporation | Oct 21, 2026 | 17 | 0 | 0 |
| WesBanco | Oct 21, 2026 | 17 | 0 | 0 |
| Wells Fargo | Oct 21, 2026 | 17 | 5 | 2 |
| Webster Bank | Oct 21, 2026 | 17 | 1 | 1 |
| WaFd, Inc. | Oct 21, 2026 | 17 | 0 | 0 |
| Visa | Oct 21, 2026 | 16 | 0 | 0 |
| Varo Bank | Oct 21, 2026 | 16 | 0 | 0 |
| Valley National Bancorp | Oct 21, 2026 | 17 | 0 | 0 |
| USAA Federal Savings Bank | Oct 21, 2026 | 17 | 0 | 0 |
| U.S. Bancorp | Oct 21, 2026 | 17 | 4 | 1 |
| Upstart | Oct 21, 2026 | 16 | 9 | 3 |
| Upgrade, Inc. | Oct 21, 2026 | 16 | 0 | 0 |
| United Community Banks | Oct 21, 2026 | 17 | 0 | 0 |
| United Bankshares | Oct 21, 2026 | 17 | 0 | 0 |
| UMB Financial | Oct 21, 2026 | 17 | 0 | 0 |
| UBS (US) | Oct 21, 2026 | 17 | 0 | 0 |
| UBS Group AG | Oct 21, 2026 | 17 | 0 | 0 |
| Trustmark Corporation | Oct 21, 2026 | 17 | 0 | 0 |
| Truist Financial | Oct 21, 2026 | 17 | 0 | 0 |
| TowneBank | Oct 21, 2026 | 17 | 0 | 0 |
| Toast | Oct 21, 2026 | 16 | 0 | 0 |
| The Bancorp, Inc. | Oct 21, 2026 | 17 | 0 | 0 |
| Texas Capital Bancshares | Oct 21, 2026 | 17 | 0 | 0 |
| TD Bank (U.S.) | Oct 21, 2026 | 15 | 0 | 0 |
Know of an AI incident at a bank or fintech that is missing, or a correction to one listed? Reply to any issue of the daily brief with a link to the coverage, or use the contact on the About page. Suggestions are checked against the same sourcing rules before anything is added, and corrections are dated on the record.
What counts as an AI incident at a bank?
An event, dated 2022 or later and publicly reported, in which an AI system caused or contributed to harm or a reversal at a bank or fintech: the institution's own AI failing, misbehaving or being withdrawn; criminals using AI (deepfakes, voice clones, AI phishing, synthetic identities) against the institution or its customers; or a governance finding such as a bias ruling, a lawsuit, a regulator's finding or AI-washing. Policies, pilots and restrictions with no harm are listed separately as responses and near-misses.
Why does the register start in 2022?
Generative AI reached banks' customers and staff from late 2022, and incident reporting about bank AI before then is sparse and mostly about older statistical models. A fixed start date keeps the counts comparable year to year; earlier events are out of scope rather than missing.
How is severity scored?
On a 1–5 scale: 1 contained, no customer harm; 2 limited harm or a policy reversal; 3 material harm or a regulatory finding; 4 large losses, many customers, or a consent order; 5 systemic, criminal, or across several institutions. Each record states the reason for its score.
What do the status labels mean?
Reported: covered by the press but not confirmed by the institution. Confirmed by the institution: the bank or fintech acknowledged it. Regulator finding and court filing: a regulator or court document describes it (a filing is an allegation until decided). Settled, resolved and disputed say how it ended. Reported, disputed and court-filing records use attributed wording such as "alleged" or "according to".
What sources does an entry need?
At least two sources from two different publishers, or at least one official or court source; every paragraph, timeline event and impact figure cites the sources behind it. Aggregator and content-farm sites, Wikipedia, Medium and LinkedIn posts are not accepted; official statements are quoted where they exist, and the institution's response is always recorded, even when it is "no public statement found".
Does a bank without an entry mean it has had no AI incident?
No. It means none has been found that meets the sourcing bar. Each bank and fintech page says which case applies: incidents on record; a dated sweep that found none ("as of <date>, no publicly reported AI incident found"); or not yet swept. 190 institutions have been swept so far.
How do I report a missing incident or a correction?
Reply to any issue of the daily brief with a link to the coverage, or use the contact on the About page. Every suggestion is checked against the same sourcing rules before it is added, and corrections are dated on the record.
Where are bank bans on ChatGPT and other precautions?
On the responses and near-misses page: LLM restrictions, paused deployments, regulator warnings and internal probes where no harm was reported. They show how banks reacted to AI risk, so they are dated and sourced, but they are not counted as incidents.
When AI goes wrong at a bank, the next morning's brief says so.
when one of these regulators moves, the next morning's brief says so · six sourced stories · 7 am PT · free
plus every regulation, bank and agent page update, the morning after · leave any morning