What modern casino cheating technology actually looks like

The short answer: the equipment needed to cheat a table game is now ordinary consumer hardware, and the prosecuted cases of the last two years keep taking the same shape: a concealed camera, an accomplice somewhere else, an earpiece. What those attacks steal is the identity of specific cards. The attack that has not yet turned up in a courtroom is the one that steals the arithmetic instead, and there is exactly one class of bet where it would be worth the trouble. Both end in the same place: someone sizing their bets to information they should not have.

The cost curve collapsed

None of the following required a laboratory. The cameras are consumer parts, the relay is a phone, and the analysis is a person sitting somewhere else.

Two things are worth separating in that list. The first three are attacks on live casino games, and the fourth is a public demonstration against equipment those games run on. The last is an indictment over underground games, where the operators controlled the room and the equipment, a much easier problem than compromising a monitored pit. What makes it relevant is that the shuffler technique had already been demonstrated publicly two years earlier. The gap between a conference talk and an allegation in a federal indictment was about two years.

What these attacks actually steal

Every case above steals the same commodity: the identity of a card someone should not be able to see. A dealer's hole card, the order of a deck, the value under a face-down community card. That payload is worth enormous money immediately (knowing one hidden card can swing a hand from a house edge to a player edge outright), and it needs almost no computation. A person with a screen and a phone is sufficient.

That is why the record looks the way it does. Nobody is building anything clever, because they do not need to.

The attack that has not shown up yet

There is a different design that gets discussed whenever the subject of machine vision and casinos comes up: several players seated at different tables, cameras in ordinary-looking glasses, automatic card recognition, and a signal to the rest of the team whenever a shoe turns favourable. It is worth taking seriously, and it is worth being precise about which parts are hard.

Recognition is not the hard part, and has not been for fifteen years. In 2009 a University of Dundee system read cards from a single overhead camera at better than 99% accuracy while using stereo imaging to measure the height of chip stacks, in real time. It did this to detect counting, from the casino's side. Open-source object detectors now do the same job on hardware that costs very little. Anyone claiming this is a future problem is a decade behind.

The hard parts are physical and economic. A composition-based edge requires seeing every card that leaves the shoe, from a seat at the table, with hands and bodies in the way. Cards missed degrade the estimate, and the estimate is the whole product. Compare that to a hole-card attack, which needs one card, once, and yields a far larger edge for far less exposure. On a main game the arithmetic simply does not favour the sophisticated approach, which is the most likely reason no prosecution has featured one.

There is one exception, and it is the important one. Some side bets pay on the concentration of specific ranks rather than on a general high–low balance. Their probability is a sum of squares across thirteen rank counts, and no single running number can represent that, a mathematical limit rather than a gap awaiting a cleverer system, as set out here. On those bets there is no skilled human alternative to be beaten. Machine assistance is not an efficiency gain; it is the only practical route in. That is the one place where the cost of a multi-table team with automatic signalling buys something no amount of talent can.

So the two threat lines (the concealed-camera relay that keeps appearing in court, and the side-bet calculator that so far has not) converge on the bet class where nobody is watching for a human tell, because there is no human tell to watch for.

Hunting the device is the losing half of the problem

In both the Sydney and the French case, the thread was pulled by a person noticing something physically odd. That works, and the staff involved deserve the credit. It is also luck-dependent, and it happens after the money has gone.

The trend runs against it. A decade ago a wearable camera was a conspicuous novelty and casinos banned it on sight. Consumer camera glasses are now visually indistinguishable from ordinary frames, which is precisely why venue bans have become difficult to enforce by observation. Every year the device gets smaller, cheaper and more normal-looking. A detection strategy whose primary sensor is a supervisor's eyesight is being asked to win a race it is structurally losing.

Physical countermeasures still matter; they are just the wrong thing to rely on alone.

What does not change is the money

Whatever the payload and whatever the hardware, stolen information only becomes revenue through the betting box. It has to be spent as a wager, at the right moment, in the right size.

That constraint is unusually strong. Card-counting analysis summarised in the Dundee work puts 70–90% of a player's advantage in bet sizing, with only the remainder coming from altered playing decisions. An attacker who suppresses their bet spread to look normal is throwing away most of what the device bought them. The signal is not incidental to the attack; it is the attack.

Which produces a useful asymmetry:

• On blackjack, a device-assisted player and a genuinely skilled counter produce a similar statistical picture. Counting without an aid is legal in most jurisdictions, and the response is procedural.
• On a rank-concentration side bet, the skilled-counter explanation does not exist. A sustained relationship between wager size and composition-derived edge points to information arriving from outside the player's head, because there is no inside-the-head method that produces it.

That second case is a different category of finding with a different response path (escalation rather than a shuffle-up), and it is measurable from records surveillance teams already keep. It needs no camera, no reader and no integration with the equipment being attacked, which matters when the equipment itself is one of the threat surfaces.

Practical response

Five things, roughly in order of cost:

1. Card delivery. Most concealed-camera attacks depend on a card being briefly visible at table level. Dealer training and removing anything that forces a card to be lifted attack the payload at its source.
2. Devices at table level. A phone lying flat on the layout is a camera pointed at the game. Policies written before periscope attachments existed are worth re-reading.
3. Equipment inspection. Exposed ports, seals and tamper evidence on shoes, shufflers and chip trays, checked on a schedule rather than after an incident.
4. A device policy that survives contact with reality. Wearable cameras can no longer be identified by looking at them; a rule that assumes otherwise gives false comfort.
5. A standing measurement of wager against edge, kept as a baseline rather than assembled reactively during an investigation. Baselines answer "is this unusual?"; investigations started from scratch rarely can.

One legal note that changes how an incident file is built: the offence generally does not turn on whether the device actually worked. Western Australia's Casino Control Act 1984 s30(3) makes it an offence to use or possess in a casino, for the purpose of fraud, cheating or stealing, any machine or equipment that "permits or facilitates, or is intended to permit or facilitate" it. That last clause is the important one: a device that failed is still inside the section. The other Australian states carry equivalent cheating and possession provisions in their own casino control legislation, and Nevada's NRS 465.075 likewise reaches possession, with intent to use, of a device designed to obtain an advantage. The UK is built differently: s42 of the Gambling Act 2005 creates an offence of cheating, or of doing anything to enable or assist another person to cheat, rather than a possession offence.

The practical consequence is that a statistical case and a recovered device support each other. The numbers establish that something happened across a period; the device establishes how. Either one alone is materially weaker.

---

EdgeTrack Lab computes the true odds of a composition-dependent side bet from what remains in the shoe and correlates that edge against each player's wagers, hand by hand, from the logs your surveillance team already keeps, with no hardware to install, no integration with gaming equipment, and no personal information required. If that is the exposure you are trying to measure, book a demo.

Sources

Every case and figure above comes from published reporting, court records, or peer-reviewed work:

• SBS News: Kazakhstan couple arrested over alleged $1m casino scam, for the Sydney amounts, dates, devices seized and how the pair were noticed
• Willy Allison, World Game Protection Conference: Technology-enabled cheating: a wake-up call for the casino industry and Cheats, hidden cameras and exposed cards, for the threat taxonomy, the €200,000 figure in the French case, and the countermeasure recommendations
• PetaPixel: Poker players are cheating with tiny hidden cameras, for the French mirror-attachment method and how the accomplice relay works
• Casino.org: Multistate baccarat scam nets $1.5 million, for the cut-card fanning technique and its scope
• Wikipedia: 2025 NBA illegal gambling prosecution, summarising the October 2025 indictment and the equipment alleged in the private games
• WIRED / IOActive research presented at Black Hat 2023, summarised by Kaspersky, for the shuffler USB attack and the internal-camera deck-order leak
• Casino Control Act 1984 (WA) s30, published at legislation.wa.gov.au, for the possession provision and its "intended to permit or facilitate" wording. Nevada's NRS 465.075 and s42 of the UK Gambling Act 2005 are on their respective legislation registers
• Krists Zutis and Jesse Hoey, Who's Counting?: Real-Time Blackjack Monitoring for Card Counting Detection, University of Dundee, ICVS 2009, for the >99% card recognition figure, real-time bet tracking, and the 70–90% share of advantage attributable to bet sizing