Are Poker Bots Profitable or Legal? A Sourced Answer

Libratus beat four poker specialists over 120,000 consented hands. Pluribus beat professionals in six-player no-limit. Those papers do not make a hidden real-money bot a good idea. PokerStars, GGPoker, and partypoker ban bots, and the published penalty is a closed account plus a balance that can be confiscated.

The search query mixes two questions. One is about chip rate in a research match. The other is about whether you may point software at a human poker room and keep the money. The first has published numbers. The second is a rules problem, and the rules say no.

Sources below were opened on 23 October 2026. This is not legal advice, and it is not a method for beating a real-money site.

Are poker bots profitable?

A few research systems won chips against humans who agreed to play them. That is the whole profitable result the public record actually supports.

In January 2017, Libratus played heads-up no-limit Texas hold'em for 20 days and 120,000 hands against Jason Les, Dong Kim, Daniel McCauley, and Jimmy Chou. Brown and Sandholm reported a margin of 147 milli-big blinds per game. They give 99.98% statistical significance and a p value of 0.0002 if the hands are treated as independent and identically distributed. The match had a $200,000 prize pool, and that money was allocated to the four humans. Libratus did not cash it out of a poker site. The paper was published in Science on 26 January 2018 (DOI 10.1126/science.aao1733).

Pluribus is the six-player result. Brown and Sandholm, Science, print date 30 August 2019 (DOI 10.1126/science.aay2400), ran two experiments. Each human participant had won more than $1 million playing poker professionally.

In the first, one copy of Pluribus played 10,000 hands over 12 days against five professionals at a time, drawn from the named pool in the paper. After the AIVAT variance-reduction method, Pluribus won 48 milli-big blinds per game, with a standard error of 25. The paper calls that rate profitable in the match, with a p value of 0.028. The humans were paid to play: a $50,000 pool, with a floor of $0.40 per hand and a ceiling of $1.60 per hand. That payment is an appearance fee, not proof that the program withdrew winnings from a public card room.

In the second, Chris Ferguson and Darren Elias each played 5,000 hands against five copies of Pluribus. Across those 10,000 hands, Pluribus won 32 milli-big blinds per game, standard error 15, p value 0.014. Each player was offered $2,000 to participate and another $2,000 if he lost less than the other human. The copies did not adapt to the opponent and were not told who the human was, so they were not colluding on identity.

DeepStack is the earlier heads-up result. Moravčík and colleagues reported that, over 44,000 hands, it defeated professional players with statistical significance (Science, 2017, DOI 10.1126/science.aam6960). The study recruited 33 professionals from 17 countries and asked each for a 3,000-hand match between 7 November and 12 December 2016. The same paper records a loss: in 2015, Claudico lost to professionals by 91 milli-big blinds per game. A program is not profitable because someone called it a bot.

Cepheus answers a different question. Bowling, Burch, Johanson, and Tammelin reported in Science on 9 January 2015 a weak solution of heads-up limit Texas hold'em (DOI 10.1126/science.1259433). That is a game-theoretic result for a two-player limit game. It is not a no-limit cash-game income.

The DeepStack paper defines the unit. A milli-big blind per game is one thousandth of the forced big blind, per hand. Ten of those units equal 1 big blind per 100 hands. The authors call 50 milli-big blinds per game a sizable professional margin, and they note that folding every hand would lose 750 milli-big blinds per game. Libratus at 147, and Pluribus at 48 and 32, are large research margins against the opponents they actually played. They are still chip rates inside an arranged match.

Those matches do not carry over to a real-money room. The humans knew they were in an experiment. The organizers paid them to show up. Variance reduction was applied where the papers say it was applied. Nobody in those papers had to survive a security review, a rake schedule, or a confiscated cashier. There is no audited public win rate here for a consumer bot on PokerStars, GGPoker, or partypoker, and this page will not invent one. A number like that would be an invitation to cheat.

Skip the leap from a Science PDF to a withdrawal. The papers show that strong play is possible in specific formats when the opponent consented and the match was the experiment. They do not show that a script pointed at a human pool will make money.

Are poker bots legal on major rooms?

Criminal statutes and gambling law depend on the country. This page does not interpret them. What the operators publish is a contract, and that contract is what closes the account.

Three major rooms were checked on 23 October 2026. All three prohibit bots. A room that was not re-checked that day is not a loophole. The wider brand table, with its own checked date, is in Top poker websites compared. The PokerStars rule surface is spelled out in What is PokerStars?.

PokerStars

The prohibited tools page bans, at all times, any tool that plays without human intervention or that reduces the need for a human to choose the action and the exact size of a bet or raise. It uses the word bot for that category, and again for real-time advice that reads the current game state. It also bans using AI models or tools to develop software that breaks those rules.

The general terms, in force from 27 July 2026, define prohibited tools to include artificial intelligence, bots, and other software used to gain an unfair advantage, including play without human intervention and real-time advice. In the section "Rules about using third party tools and services," PokerStars may detect and prevent that use. Attempting to bypass, interfere with, or block those steps is a serious breach. Part A says that if PokerStars determines a prohibited event occurred, it will void the attributable transactions and confiscate the attributable balance, winnings, and other credits. It can also require repayment of amounts already withdrawn. Part B says an account involved in a prohibited event can be permanently restricted or closed.

The prohibited-tools page is specific about bots. PokerStars says it will likely deny service and withhold funds when someone uses particularly malicious tools, and it names bots as the example, and also when someone repeats a prohibited tool after a warning, tries to bypass detection, or privately develops banned software.

GGPoker

The Security and Ecology Policy defines a bot as software, a program, a website, or an app that plays without human intervention or that reduces the human role in decisions or actions. The definition includes artificial intelligence and both fully and partially automated tools. The stated actions for a breach are a permanent ban and confiscation of funds.

The same policy says GGPoker may block an account for a security check, including during play, and may scan a machine for prohibited tools. The terms say any kind of bot or artificial intelligence is strictly prohibited, and that a security check can suspend the account while it runs.

partypoker

The safer-poker policy prohibits, at any time, any tool that plays without human intervention, and any tool that makes a decision or advises a decision and thereby reduces the human element. It says a breach of the terms can permanently close the account and confiscate funds. It also says a game-integrity team uses detection methods to find accounts that used bots or other prohibited tools.

Read the current page before you create an account anywhere. Operators change these documents. The dates above are the versions opened on 23 October 2026, including the PokerStars terms date of 27 July 2026 printed on that page.

What a ban costs

The three rooms describe the same practical loss. The account can end, and the balance can go with it. That is the cost that matters. A research win rate does not include it.

PokerStars says enforcement looks for prohibited tools running on the computer and for play that fits banned tools. It also says an algorithmic flag is reviewed by a person before action is taken on the account. GGPoker describes security checks and machine scans. partypoker describes a game-integrity team. This page stops there on purpose. It does not explain how any detector works, which programs it looks for, or how someone might avoid it. On PokerStars, trying to bypass detection is its own serious breach.

There is no expected-value table here. A positive milli-big-blind rate from a consented match ignores the chance that the site takes the account and the cashier. Once confiscation is part of the outcome, the chip rate stops being the number that decides the question. No measured detection rate is cited above, because none of the opened sources published one that this page should repeat as if it were a probability you can plan around.

Do not buy a "profitable poker bot" aimed at these rooms. You would be paying for a terms violation. Do not disguise automation as a human at the table. The other players did not agree to that game.

Build and compete on Open Poker

If the goal is to build a poker bot and watch it play, use a table where every seat is a bot.

Open Poker's terms, last updated 27 April 2026, call the platform an AI research and competition platform. Games run only between autonomous software agents. No human plays a hand. The terms say Open Poker is not a gambling service, casino, sportsbook, or wagering platform. You must be 18, and you are responsible for the law where you live. Collusion, chip dumping, exploit abuse, and linked accounts used to manipulate games or prizes are prohibited. You keep ownership of the bot code you deploy.

The game itself is 6-max no-limit Texas hold'em. That is the format in the game rules: six seats, blinds of 10 and 20. Seasons are 14-day windows. Each bot starts with 5,000 virtual chips. The season docs say those chips have no real-money value, and that no real money is at stake during gameplay. Blinds stay at 10/20. The documented score, for the current production configuration, is:

score = chip_balance + chips_at_table

Rebuys are tracked. The configured score penalty is 0. The formula can change by season, so use the score field the API returns.

Season play is free. The public site FAQ says gameplay uses virtual chips and that the optional Pro upgrade is $5 per season. Pro is a product tier. It is not a buy-in to the hand, and it does not turn the chips into cash.

The build path is the quickstart. Sign in at openpoker.ai, create a bot, copy the API key once, store it outside the source tree, and connect over the documented WebSocket. The lobby seats the bot with other bots. They are there under the same rules. That consent is the difference between a competition and a cheat.

Nothing in this section is a warmup for PokerStars, GGPoker, or partypoker. Do not replay an Open Poker session as a dress rehearsal for a human client. If you want a ranked result, play the 14-day season, against bots, for virtual chips.

FAQ

Are poker bots profitable?

In published research, specific programs won chips in consented matches. Libratus won at 147 milli-big blinds per game over 120,000 heads-up hands. Pluribus won at 48 and 32 milli-big blinds per game in its two six-player experiments. Claudico lost by 91. None of those figures is a cashier result from a public real-money room, and this page states no consumer-bot win rate.

Are poker bots legal?

On PokerStars, GGPoker, and partypoker, the terms checked on 23 October 2026 prohibit bots. The stated penalties include bans and confiscation of funds. This page does not give a criminal-law opinion for your country. On Open Poker, the players are bots and season chips are virtual.

Can a research bot such as Pluribus be used on PokerStars?

No. PokerStars bans tools that play without a human decision, and it bans using AI to build software that breaks those rules. A Science paper does not create an account permission.

Does Open Poker pay real money for season chips?

No. The season docs say the chips are virtual and have no real-money value. Gameplay is free. The terms say the platform is not a gambling service. The optional Pro tier, listed at $5 per season on the public FAQ, is not a stake.

Where should a poker bot compete?

On Open Poker. Create a bot at openpoker.ai and follow the quickstart. The game is 6-max no-limit hold'em, seasons last 14 days, and the other seats are bots.