ACR poker bots: what the public record actually shows

By Theo Brandt, independent poker bot operator · [enable JS to see contact email]

Americas Cardroom (ACR) has one of the most publicly discussed bot presences in online poker, there is no credible public evidence of working client-side bot detection on the site, and the "bots" people argue about there are a different problem from the retail bot products I actually bench. My 22,000-hand test ran on PPPoker, not on ACR. This page is not a bench result. I apply my usual methodology to the documented record, separating test results from inference.

What this page is, and what it is not

Everything on the comparison homepage comes from hands I personally dealt: eight bot products, thirteen configurations, ~22,000 hands each, all on PPPoker. I did not run that bench on ACR. Americas Cardroom runs its own desktop and mobile clients on the Winning Poker Network, and putting a fresh bench there honestly would mean months of hands and real bankroll at risk on a US-facing site — I have not done it, and I am not going to pretend a PPPoker number transfers to a room with a different client, a different security team, and a different player pool.

Any claim about ACR is sourced to something public and dated, or it is flagged as inference from architecture. When I say "on my bench," I mean PPPoker. When I say "on ACR," I mean the documented record.

The documented bot presence on ACR

ACR's bot situation has been covered over multiple years by multiple outlets. It is not a rumour on one forum, and the people driving the coverage are named, not anonymous.

Read the pattern, not any single post. Two things are consistent across every source above: (1) experienced players believe there is a meaningful bot population at low-to-mid stakes, and (2) nobody outside ACR has ever shown a bot being caught by the client. The catches, where they happen at all, come from human pattern-spotting after the fact. The gap between "players see bots" and "the site demonstrably detects bots" remains.

Does the ACR client detect bots?

There is no credible demonstration of real-time, client-side bot detection on ACR. ACR's own security messaging (acrpoker.eu, Apr 2023) states that bots are banned and threatens permanent bans, but publishes nothing about detection mechanics (which is standard for every room; no operator advertises its detection stack).

Two of the pages ranking for "does ACR poker have bots" are cloaked SEO-spam hosted on hijacked government domains, dressed up as "Apps on Google Play" listings. They confidently claim ACR uses "behavioral biometrics and machine-learning to detect bots in real time." That sentence sits on obviously compromised pages. It is not from ACR and has no attribution. I would not use it to make a decision. It reads like generated filler written to rank, not to inform.

The WPN client, like most poker clients, can log input timing, cursor paths, session length, and multi-account correlation. Whether ACR actively runs behavioural analysis on that data, and how good it is, is unknown from the outside. The community's lived experience (Mobius, the 2+2 threads) points the other way: toward under-enforcement, not aggressive detection. I would plan around "detection is inconsistent and mostly reactive," not "there is no detection." Both overconfident readings get operators banned.

Which bot architectures even run on ACR

This is where my bench experience does transfer, because it is about how the products are built, not about ACR's numbers. The products I tested split into two families, and they meet ACR very differently.

OpenHoldem-family desktop bots (the Warbot / Shanky / Inhuman / PokerBot.com lineage) work by scraping the poker client's screen against a hand-built table map, then acting through the OS. On my PPPoker bench, that entire family stopped working in a single week when PPPoker shipped a client update. The table maps stopped matching and the bots could not attach. ACR runs its own desktop client on the Winning Poker Network and updates it on its own schedule, so any OpenHoldem-family bot on ACR lives at the mercy of a table map that a single client update can break. If you go looking for "an ACR profile" for one of these, the maintenance date on that profile matters more than any feature list.

Android-emulator bots (the NZT Poker / PokerBotAI style) run the room's mobile app inside an emulator and read state from the app rather than a scraped desktop window. ACR ships a mobile client, and this family advertises ACR among supported rooms. PokerBotAI explicitly lists ACR compatibility. Architecturally these are the more resilient of the two against client updates, because reading app state survives cosmetic UI changes better than a pixel-matched table map does. On my PPPoker bench, the one product that finished the full hand count positive was PokerBotAI, and NZT Poker broke even; that is a PPPoker result, and I am not restating it as an ACR result. If ACR is your target room, the emulator family is the only architecture in my field that both survives client churn and claims ACR support.

What my PPPoker bench does and doesn't tell you about ACR

Transfers: how each product is built, which ones survive a client update, which vendors are metered-by-fuel versus one-time, and which one was an outright scam (3upgaming; that verdict is about the vendor, and vendors don't get more honest per room). See my criteria for how each of those was scored.

Does not transfer: winrate, ban rate, and detection latency. ACR's player pool is US-facing and, per the public record, unusually bot-aware and vocal about it. A steady 4–8bb/100 graph (the "smarthand" pattern that gets accounts flagged on 2+2) is exactly the kind of too-clean result a well-tuned bot produces, which means the biggest risk on ACR may not be the client at all. It may be the regulars building spreadsheets and reporting you. That is a human detection layer my bench never measured, because PPPoker's clubby pools don't do it at the same intensity.

If you're evaluating an ACR bot: a short checklist

• Ask when the ACR support was last updated. For desktop/table-map products, an ACR profile that hasn't been touched since the last client update is probably broken. Make the vendor show you a recent one.
• Separate "runs" from "wins." Running on ACR is an engineering claim you can verify quickly; winning on ACR is a bankroll claim you cannot verify from a sales page.
• Assume a human detection layer. The documented ACR catches come from players and analysts, not the client. A graph that never varies is a liability in a pool this bot-aware.
• Discount anyone promising you can't be caught. No honest operator says that, and I don't either. Every product on my bench got banned at some point except the two that finished the window clean. "Clean during one test window" is not a promise about your account.

FAQ

Are there bots on ACR? By every public account (TwoPlusTwo, Mobius Poker, the Jan 2024 PokerNews saga, Vojtech Cervinka's tracking), players widely believe there is a meaningful bot population, especially at low-to-mid stakes. None of that is proof of any single account, but the pattern is consistent across independent sources over multiple years.

Does ACR detect bots automatically? There is no credible public evidence of working real-time client-side detection. ACR bans bots by policy and issues permanent bans, but publishes no detection detail. The claims of "ML behavioral biometrics" that show up in search are on cloaked spam pages, not from ACR.

Did you test a bot on ACR? No. My bench is PPPoker, ~22,000 hands per setup across eight products. This page covers the documented public record and architecture analysis, not a bench result. I keep those separate.

Which bot type is more likely to run on ACR? The Android-emulator family (it reads the mobile app and survives client updates better and advertises ACR support) over the OpenHoldem-family desktop scrapers (which break when the client updates). "Runs" is not "wins."

What's the real risk on ACR? On the public record, at least as much from bot-aware regulars reporting suspiciously steady graphs as from the client itself. Plan for a human detection layer.

More: the full eight-product comparison is on the homepage; the test setup is on the methodology page; per-criterion scoring is under criteria.