AI is already inside the casino: tables that see, cameras that recognize, and the risk that all games will look the same

Artificial intelligence is no longer just a buzzword. In the gaming industry, it has been appearing in corporate presentations for years, but only in recent months have use cases that truly move the needle begun to emerge. And they are far less spectacular—and far more profitable—than what marketing promises.

The honest summary is this: AI is not replacing the game designer or the croupier. It is accelerating the work that teams were already doing, and it is placing eyes where before there was only human memory.

The quote comes from Bob Serr, Chief Technology Officer of Aristocrat Leisure, and summarizes the stance of the world's largest slot machine manufacturer. In his recent investor session, Serr explained that the company focuses its use of AI on three fronts: boosting creativity, shortening time to market, and extracting better insights from its own data.

The journey began in late 2022, when ChatGPT became public and Aristocrat launched a phase of experimentation with generative AI. In 2024, it deployed tools such as Google Gemini, GitHub, and Copilot across its development teams. In 2025, it moved from “using tools” to redesigning complete workflows with autonomous agents.

The figure that best illustrates this comes from Product Madness, its social platform: between 2021 and 2025, it went from around 8,000 to 14,000 creative assets per year—a 75% increase—with fewer artists and designers on staff.

Ian Hughes, Chief Revenue Officer of GLI and CEO of GLI Australia—the laboratory that certifies a large share of the world's machines—points to the less glamorous and more real side of the matter: the code.

“Some studies have found that up to 80% or 90% of a developer's codebase can now be written by AI,” Hughes explains. With an important nuance: “That doesn't mean designing games. The design, architecture, and creative elements are still done by people. But low-level, mechanical code can now be delegated.”

The second major use is prototyping. Previously, validating an idea cost months or years of development with no guarantee that it would reach the market. Today, a functional mockup is built and taken to the expo. “We can put things in front of the customer, take them to a trade show, and validate the idea before assigning those precious development hours to it,” notes Joe Mayer, Systems Sales & Operations Manager at Konami Australia.

On top of all that, agentic AI has arrived. Hughes describes it with a domestic analogy: “It's like the general contractor who says 'I have this job to do' and has agents working for him, each very specific. You end up with a better solution because each agent performs a single task.”

This is where AI becomes tangible. Smart tables are already deployed or being rolled out across all Macau casinos, at Solaire and Newport World Resorts in Manila, at Marina Bay Sands in Singapore, at SkyCity in Auckland, and in Australia at The Star Sydney and Crown Melbourne.

The principle is simple to explain and difficult to execute: they combine RFID in the chips with computer vision to track the movement of every bet, attribute it to a specific player, and ensure that exactly what is owed is paid out or collected.

The benefits go far beyond eliminating dealer error: fraud detection, identification of advantage play, and—the one that matters most to the operator—a true assessment of the player. Players are no longer rated based on what the supervisor remembers they wagered, but on what they actually bet, hand by hand.

The two major manufacturers in Asia-Pacific have taken different paths. Walker Digital Table Systems uses AI primarily for data analytics through applications. Angel Group relies on a “hybrid” system: two cameras on each side of the table that automatically capture the position, quantity, and type of chips in each stack, record player actions, and verify that the number of RFID chips in each betting spot matches what the camera sees.

The rollout began with baccarat, but blackjack and roulette are already arriving. The ultimate goal is total coverage: every bet on every table, recorded.

The great irony is that facial recognition did not enter Macau through concessionaires, but through the government: in 2017 it mandated the installation of facial KYC systems at all ATMs, starting with those in casinos, to control capital outflow from UnionPay cards issued in mainland China.

Two years later, when operators began adopting it, the regulator ruled that it should not be used to identify high-value players. A senior concessionaire executive sums it up bluntly: “The regulatory environment in which we operate undeniably limits the application of AI computer vision. Privacy regulations on facial recognition are not clearly defined.”

In Australia, the exact opposite occurs: the technology is not tolerated, it is required. Crown Resorts and Star Entertainment use it to detect self-excluded individuals and bar them from entering, and the New South Wales government is considering making it mandatory in every gaming room of pubs and clubs—where the vast majority of the country's 185,000 machines are located—to support the state exclusion register.

Konami is working on a friendlier variant: SYNK Vision, a “facial authentication” system that recognizes members as soon as they sit down and registers new players as guests. The technical key lies at the edge: “With the optimized algorithms we have now, we can process everything at the edge, so the face embedding never leaves the camera,” explains Mayer. Better security and lower cost, all in one move.

The first is data security. Mayer puts it bluntly: “For all the operational value you can get from AI, a single major data breach can destroy that value overnight.” Konami limits its use of AI to proof-of-concept trials and approved internal platforms: “Nothing generated by AI ever gets anywhere near production.”

The second is intellectual property, and it is more uncomfortable than it seems. “IP protection does not exist for AI-generated content,” warns Hughes. “If you use agentic AI to create a character and a human didn't create it, there is no copyright on that.” Even worse: since generative AI learns from existing patterns, the design might inadvertently incorporate pieces of someone else's characters. “It could have used some element from The Flintstones without you even realizing it.”

The third is the quietest: the death of original content. “AI works by looking at what already exists and recombining it,” Hughes insists. “If there are no game designers with fresh ideas and we just regurgitate, everything that launches will be a deviation or the fourth iteration of something that already exists. That would be sad.”

The first: the jackpot that breathes with the floor. A mystery jackpot whose hit range is adjusted by a predictive model based on actual occupancy, the day, the weather, or city events, always within a certified and published mathematical range. The theoretical return is not altered; the excitement is distributed where and when the floor needs it. The floor ceases to have a flat 3:00 p.m. curve.

The second: anomaly detection within the jackpot network itself. A progressive system sees the wagering pattern of each connected machine. A model that learns the normal footprint of each game detects physical tampering, collusion, or malfunctions before any camera does, because it does not interpret images: it interprets money in motion.

The third: responsible personalization without cameras. Facial recognition is not needed to know that a patron has been seated for four hours. The jackpot system already knows this through the flow of credits. That same signal, handled with sound judgment, serves responsible gaming—a suggested break, an offered limit—far better than a photo in a database, and without the privacy alarms that have Macau paralyzed.

The fourth: audited generative narrative. Here it is indeed wise to heed GLI's warning regarding copyright. Our approach is not for AI to invent characters, but to modulate the staging of an already determined prize: the timing of the announcement, the language, the music, the payout ceremony. The certified mathematics remain untouched; what the machine learns is how the prize is told, not how much it pays.

That is the fundamental difference. The industry is using AI to cut costs. We believe that AI in a land-based casino is only justified when the player experiences it as something good that happened to them, not as a camera watching them.

It is worth being explicit, because in this sector good ideas die in the regulator's room. No AI model must alter the RTP, the prize probability, or the certified mechanics of a game. Period.

AI operates at the presentation, operational, and analytics layer. The randomness engine remains a certified and auditable random number generator. Everything an intelligent system decides must be recorded in an immutable log that a laboratory like GLI or BMM can review line by line.

With that boundary respected, the progressive jackpot becomes the most natural AI infrastructure a land-based casino has. It is already installed, already connected to everything, and already has the player's trust. It just needed to be thought of that way.

The original report closes with a sensible conclusion: the greatest value of AI lies not in replacing people's work, but in amplifying it. We subscribe to every word, and add a coda from the Latin American operator's perspective.

Amplifying people's work is fine. Amplifying the player experience is what fills a casino floor. And in a land-based casino, the only technology that speaks to all players at once, every day, is the jackpot. That is where artificial intelligence needs to reach.