BRENNEN

Industries / Retail, hospitality and personal services / Hospitality and tourism / Gaming and casinos

Table games

Table games are the card, dice and wheel games that staff deal at a table, such as blackjack, baccarat, roulette and craps.

6

Milestones achieved, from 1961 to 2017.

1

To watch for: announced, not yet confirmed outside.

2

Open deliverables. Nobody has shipped them.

The road

AchievedWatch forOpen

Achieved

  1. 1961

    Card counting is proven

    Edward Thorp, a mathematician at MIT, published "A favorable strategy for twenty-one" in the Proceedings of the National Academy of Sciences in January 1961. He had used an IBM 704 computer to show that a player who tracks the cards already dealt can beat blackjack, and his 1962 book Beat the Dealer spread the method. The UC Irvine Libraries exhibit on Thorp says the Nevada Resort Hotel Association restricted splitting and doubling and reshuffled sooner from 1 April 1964, then restored the old rules when people stopped playing. Casinos also moved from single decks to 4- and 6-deck shoes, which they expected would defeat counters.

    LessonRule changes aimed at a few skilled players drove away ordinary customers too, so casinos turned to the shoe and the shuffle instead.

    pmc.ncbi.nlm.nih.govexhibits.lib.uci.eduexhibits.lib.uci.edu

    Credit

    Edward O. Thorp

  2. 1982

    Counters cannot be barred

    Resorts International in Atlantic City had kept Kenneth Uston off its blackjack tables since 30 January 1979 because he counted cards. On 5 May 1982 the New Jersey Supreme Court held that state regulation of casinos was so complete that only the Casino Control Commission, not a casino, had the power to exclude card counters.

    LessonIn a market where the regulator writes the game rules, a casino could not ban a legal skill on its own.

    leagle.comgamblingandthelaw.com

    Credit

    New Jersey Supreme Court; Kenneth Uston

  3. 2002

    Macau ends its monopoly

    STDM held Macau's casino monopoly from 1962. When it expired at the end of 2001, the government ran a public tender and on 8 February 2002 granted concessions to SJM, Galaxy and Wynn Resorts (Macao). Macau's casinos now run mostly on one table game: the regulator's figures give baccarat MOP 210.6 billion of MOP 247.4 billion in casino gross revenue in 2025, about 85%, against MOP 2.7 billion for blackjack.

    dicj.gov.modicj.gov.mo

    Credit

    Macao SAR government; figures from the Gaming Inspection and Coordination Bureau (DICJ)

  4. 2005

    Chips that carry an ID

    Wynn Las Vegas opened in 2005 with betting chips that hold RFID tags, small radio tags that report a serial number to a reader, so that readers around the property could spot fake chips. The Las Vegas Advisor wrote in 2012 that Wynn was the first casino to use them and that they had become standard. When an armed robber took chips from a Bellagio craps table in December 2010, the casino switched to its reserve set of chips and deactivated the stolen ones, worth about $1.5 million.

    LessonA chip that the casino can switch off is worth nothing to a thief.

    metzdowd.comlasvegasadvisor.com

    Credit

    Wynn Las Vegas (Rick Doptis, vice president of table games)

  5. 2013

    A shuffler fails a test

    A maker of casino equipment asked Persi Diaconis, Jason Fulman and Susan Holmes to check a new shuffling machine with ten shelves that was meant to mix a deck in one pass. The engineers' own tests had looked fine, but the mathematicians showed that a player who knew the design could guess about 9.5 cards of a 52-card deck correctly, against about 4.5 for a well-shuffled deck. The maker adopted their advice, and the study appeared in the Annals of Applied Probability in 2013.

    LessonSimple checks chosen by a machine's own designers can miss a pattern that a targeted test finds.

    arxiv.org

    Credit

    Persi Diaconis and Susan Holmes (Stanford University), Jason Fulman (University of Southern California)

  6. 2017

    Edge sorting ruled cheating

    In August 2012 the poker player Phil Ivey won £7.7 million at Punto Banco, a form of baccarat, at Crockfords casino in London. He had a croupier turn chosen cards so that tiny printing differences on their long edges showed high cards from low, and the casino's shuffling machine kept the cards facing the same way. Crockfords refused to pay, and on 25 October 2017 the UK Supreme Court unanimously held that this was cheating. The same judgment changed the test for dishonesty in English criminal law by dropping the subjective second step of the 1982 Ghosh test.

    LessonFlaws in card printing became a security risk once a player could control which way the cards faced.

    supremecourt.uk

    Credit

    UK Supreme Court (judgment by Lord Hughes); Genting Casinos

Watch for

  1. Smart tables read every bet

    Angel Group says its smart table system, which combines camera-based AI with RFID chips, is installed on thousands of baccarat tables worldwide and recognizes chips and their exact positions automatically. It says its upgraded blackjack table "recognizes dealer movements directly".

    Achieved whena regulator, test lab or casino not paid by Angel publishes measured error rates for chip, bet and dealer-action recognition on live tables.

    g2easia.com

    Claimed by

    Angel Group, 2026 (exhibitor feature for G2E Asia 2026)

Open

  1. 01

    Open test for table cameras

    An openly licensed video dataset of blackjack and baccarat rounds dealt at home with toy chips and plastic cards, filmed from overhead and from the chip tray. Every bet, card and payout is labelled, and some payouts are deliberately wrong. Add a scoring script that reports how many wrong payouts a system catches and how many false alarms it raises per hour of play, with baseline scores for open models.

    The brief
    Credit · unclaimed
  2. 02

    Score card backs for asymmetry

    A scanner test that measures how much each card's back differs from the same back turned 180 degrees, especially along the long edges, so that a casino or card maker can screen decks before use. Calibrate it on decks the builder has printed with planted edge differences of known size. Publish the method and code, and share any results on commercial decks only with their makers, so the test does not become a shopping list for cheats.

    The brief
    Credit · unclaimed

Open 01 · the brief

Open test for table cameras

Credit · unclaimed

The first to publish is credited on this line, and the deliverable moves from Open to Achieved.

What to build

An openly licensed video dataset of blackjack and baccarat rounds dealt at home with toy chips and plastic cards, filmed from overhead and from the chip tray. Every bet, card and payout is labelled, and some payouts are deliberately wrong. Add a scoring script that reports how many wrong payouts a system catches and how many false alarms it raises per hour of play, with baseline scores for open models.

Why it is still open

Vendors describe camera systems that recognize chips, dealer moves and errors, but this research found no independent accuracy figures for them. The closest published work, DeepGamble (IEEE ICMLA 2020), reported about 95% accuracy for bets and 97% for cards from 900 training examples filmed on its own cameras, gives no sign that the data was released, and found that dim lighting confused red and black chips. Willy Allison, who moderated a panel on automated surveillance at the 2026 World Game Protection Conference, wrote that "Casino table games are not effectively being monitored by humans."

arxiv.orgcdcgaming.com

Built with what exists

Two or three consumer 4K cameras on fixed mounts, a felt layout, toy chips and plastic cards; CVAT or Label Studio for labelling; Ultralytics YOLO and Meta's SAM 2 for baselines; Python.

  1. First step

    Film 100 rounds of blackjack on a kitchen table with one fixed overhead camera, label them in CVAT, and plant 10 wrong payouts.

  2. Done looks like

    At least 2,000 dealt rounds under three lighting levels, with at least 200 planted payout errors, released under an open licence with a public leaderboard; baseline catch rates and false alarms per hour published with confidence intervals.

Open 02 · the brief

Score card backs for asymmetry

Credit · unclaimed

The first to publish is credited on this line, and the deliverable moves from Open to Achieved.

What to build

A scanner test that measures how much each card's back differs from the same back turned 180 degrees, especially along the long edges, so that a casino or card maker can screen decks before use. Calibrate it on decks the builder has printed with planted edge differences of known size. Publish the method and code, and share any results on commercial decks only with their makers, so the test does not become a shopping list for cheats.

Why it is still open

In the Ivey case, differences on the cards' long edges small enough to escape a top London casino were enough to sort high cards from low. This research found no published method or public results for measuring how symmetric a deck's backs are; the only related open-source project found on GitHub was a simulator of edge sorting, not a test of cards.

supremecourt.uk

Built with what exists

A flatbed scanner at 1,200 dots per inch; a print shop that prints custom card decks; Python with OpenCV and scikit-image to align each scan with its rotated copy and map the differences.

  1. First step

    Scan the backs of two decks of the same brand at 1,200 dpi and compute the rotated-difference score for all 104 cards.

  2. Done looks like

    On 10 custom-printed decks with planted edge differences between 0.1 mm and 1 mm, and 10 unaltered control decks, the score flags every planted deck above a stated size and no control deck; three scans of each deck agree within a stated tolerance; and the scores are compared with a blind panel of 5 people judging the same decks.