Available is not collected

The missed signal: the “infinite money glitch”

On August 29, 2024, according to a lawsuit a large US bank later filed, a man in a mask deposited a check for 335,000 dollars at one of its ATMs in Texas. Most of the money was withdrawn before the check came back. On September 4, six days after the deposit, the bank it was supposedly drawn on rejected it as counterfeit. [1][2]

Over the Labor Day weekend that followed, videos on TikTok and X told viewers to do the same: deposit a large check at the bank’s ATM and take out the part made available right away. NBC reported that the first known mention was a post on X showing a balance over 80,000 dollars. [3] The internet called it the infinite money glitch.

The bank did not. On September 3 it told reporters: “Regardless of what you see online, depositing a fraudulent check and withdrawing the funds from your account is fraud, plain and simple.” [4] On October 28 it filed four civil suits in federal courts in three states. Banking Dive put the total sought at more than 660,000 dollars. [1][5] In one Georgia case described by CNBC, a check for 73,000 dollars was followed by 82,500 dollars in cash withdrawals at two branches. [6] By April 2025, CNBC reported, the bank had sent repayment letters to more than 1,000 customers. [6]

Nothing here was a bug in the usual sense. Under Regulation CC, a bank must make part of a check deposit available by the next business day. [7] The rule defines “available for withdrawal” as usable like funds that have been collected, not merely deposited. [8] Usable is not collected. Between those two moments the bank is extending credit against a piece of paper, with no credit check.

Normally that is fine. Most checks are good, and a bad one comes back to an account that can cover it. Here the accounts could not cover it, the cash was already gone, and many people tried the same move in the same few days.

Our verdict: late. CBS reported that the bank patched it within days. [1] In the cases described in court, the cash had left before the checks came back. [2][6] The public record does not say how or when the bank detected it. [5]

The control: score the withdrawal, not just the deposit

The first detectable moment was not the viral weekend. It was the first withdrawal right after a first-of-its-kind deposit, before the check could come back. This part is our analysis, not public record.

If you offer instant availability, pay advances or instant payouts, you are running the same experiment at a different size. Three signals would have flagged it:

1. Account level. A large deposit, unlike anything the account has done, followed within hours by a withdrawal up to the available amount. On its own, one customer having a strange day.
2. Population level. The same deposit then withdraw sequence spiking across many unrelated accounts in the same few days. Unrelated accounts do not learn the same move on the same weekend. When they do, someone taught them.
3. Outside the bank. That spike lining up with a public trend that names your brand and shows the steps.

The control: put the decision on the withdrawal, and give it four inputs. Deposit size against the account’s own history. Minutes since the deposit. Share of the available funds being taken. And how many other accounts did the same thing this hour. The first three catch signal one. The fourth is the only one that sees signal two, and it is the one most often left out.

The trade-off: holding a withdrawal annoys a good customer who really did deposit a large check. Route the hold to review with its reasons attached, rather than blocking, and the cost is a short delay, not a lost customer.

Signal two is also hard to see from inside one institution, because a cluster spreads across many. That is the case for fraud patterns that travel between institutions while no institution’s data reaches another.

The number: over 680,000

Financial institutions filed over 680,000 suspicious activity reports related to check fraud with FinCEN in 2022, up from over 350,000 in 2021, as cited in FinCEN’s February 2023 alert on mail theft related check fraud. [9] These counts include every report that flagged check fraud, not only mail theft.

Sources

Headlines are quoted with the bank’s name replaced by a bracketed description.

1. CBS News, 2024-10-29, “[A large US bank] sues customers it says took advantage of ‘infinite money’ glitch”: read the source
2. American Banker, 2024-10-29, “[A large US bank] sues 4 customers over TikTok check fraud ‘glitch’”: read the source
3. NBC News, 2024-09-03, “[A large US bank] says it is aware of viral ‘glitch’ inviting people to commit check fraud”: read the source
4. Mashable via Yahoo Tech, 2024-09-03, “TikTok viral [bank] money glitch is just bank fraud. Don’t do it.”: read the source
5. Banking Dive, 2024-10-29, “[A large US bank] sues customers who exploited viral ‘glitch’”: read the source
6. CNBC via a local NBC station, 2025-04-16, “[A large US bank] sues more customers who allegedly stole cash in ‘infinite money glitch’”: read the source
7. 12 CFR 229.10, Next-day availability (Cornell LII): read the source
8. 12 CFR 229.2, Definitions (Cornell LII): read the source
9. Orrick InfoBytes, 2023-03-03, “FinCEN warns financial institutions of surge in mail theft-related check fraud” (summarizing FinCEN alert FIN-2023-Alert003): read the source

---

A note from Amir. I spent years building fraud models at NICE Actimize, for large banks. I have no inside knowledge of this case; everything above comes from public reporting and court filings. What stays with me is how ordinary each deposit looked on its own. The signal was never in one account. It was in the crowd. That is the problem I worked on at NICE Actimize, and it is why every issue of Signal Log ends on a control you can build, not a warning.

Amir Shachar