Artificial intelligence has already entered gaming, but not in the form most players were promised. The current wave is not dominated by endlessly intelligent non-player characters, self-building open worlds, or game directors that rewrite entire campaigns around each player. Most companies are starting with less glamorous work. They are generating code, automating testing, detecting fraud, searching internal documents, assisting customer-service teams, and analyzing player behavior.
That distinction matters because the newly released State of AI in Gaming 2026 report from the University of Nevada, Las Vegas International Gaming Institute and KPMG offers one of the clearest pictures yet of where AI adoption stands. It also exposes a terminology problem that gaming communities should understand from the start.
The report is primarily about the regulated gambling industry, including online betting operators, casino businesses, gaming technology suppliers, slot developers, and related platforms. It is not a census of traditional PC, console, and mobile video game studios. Still, its findings reveal patterns that closely resemble what is happening across software-driven game development. AI adoption is spreading quickly, while strategy, staffing, governance, and proof of financial value remain far behind the sales pitch. The robots have arrived. Management is still filling out the paperwork.
The Report Measures Ambition More Than Mastery
UNLV’s AI Research Hub produced the report with KPMG using survey responses from 83 gambling companies and 113 regulators worldwide. The researchers also examined 15 years of academic publications, conference programs, patent filings, industry reporting, and expert commentary. This is intended to become an annual benchmark rather than a one-time snapshot.
The report’s AI Maturity Index gives the industry an overall score of 45 out of 100. Strategy performs better at 57, while governance falls to only 30. In plain language, companies are much better at announcing AI plans than building the infrastructure, expertise, oversight, and measurement systems needed to support them.
That result should sound familiar to anyone who has watched video game publishers discuss new technology. The industry has repeatedly raced toward cloud gaming, blockchain economies, virtual reality, live-service production, and user-generated content before deciding what success is supposed to look like. AI is following the same pattern, only faster.
Executives see an opportunity to cut production costs, process more data, release content faster, and handle growing workloads without expanding payroll at the same rate. Development teams see a mixed bag. Some tools remove repetitive work. Others create unreliable output that must be inspected, corrected, documented, and legally cleared. The difference between an AI demonstration and a stable production system remains enormous.
Code Generation Is the Real Breakout Use Case
Generative AI is already present in 81.5 percent of the companies surveyed. Conversational AI follows at 66.7 percent, predictive AI at 60.5 percent, computer vision at 38.3 percent, and AI agents or reasoning systems at 32.1 percent.
Within the companies using generative AI, text generation leads at 92.4 percent. Code generation is close behind at 87.9 percent. Image generation reaches 62.1 percent, while video generation sits at 36.4 percent and voice or music generation at 25.8 percent. Those figures cut through much of the public noise. AI-generated concept art may attract angry social-media threads, but software development is where companies are finding the most immediate use.
The report identifies code writing, test-case creation, technical documentation, internal web development, proprietary software development kits, data-platform work, and operational support among the current applications. Software development and testing alone represent 10.8 percent of the report’s total weighted AI activity, making that category one of the largest individual use cases measured.
For conventional game studios, the same tools can assist with editor scripts, build automation, unit tests, asset validation, localization pipelines, database queries, crash-log analysis, and repetitive gameplay scripting. None of that requires an AI to invent the game.
That may be the most realistic near-term model. Developers keep control of architecture, design, performance targets, and final review. Machine-learning systems handle narrow tasks that consume time but do not deserve weeks of human repetition.
The danger begins when management mistakes faster code generation for faster software production. Generated code still has to be reviewed for security, performance, licensing concerns, compatibility, and maintainability. A tool can produce a working function in seconds and still create technical debt that survives for years.
Technology and Product Work Lead the Adoption Race
The UNLV and KPMG data divide company activity into five broad areas. Technology and security lead with 24.5 percent of weighted activity. Product development and innovation follow at 24 percent. Business operations account for 18.7 percent, customer-facing functions reach 18.2 percent, and risk and compliance trail at 14.6 percent. Nearly half of all recorded activity is concentrated in technology, security, and product creation.
Within technology and security, companies reported AI work in software development, testing, cybersecurity, incident management, and development operations. Product activity includes game development and sports-betting systems. Customer-facing work includes support automation, player acquisition, retention, and customer-relationship systems. Risk applications include fraud detection, payment monitoring, anti-money-laundering controls, player safety, and game integrity.
Online companies report more AI activity than land-based businesses, particularly in customer-facing systems and product development. Online operators scored 54 on overall maturity compared with 39 for land-based operators. Their systems are already digital, their user activity is easier to collect, and their products can often be updated without replacing physical equipment across hundreds of locations.
Traditional game publishers have the same built-in advantage. Live-service platforms already collect telemetry, matchmaking data, crash reports, purchasing histories, moderation records, support tickets, and behavioral signals. That data can feed forecasting systems, fraud detection, retention models, automated moderation, and quality-assurance tools.
Older studios face a different problem. Decades of proprietary engines, undocumented tools, disconnected databases, outsourced pipelines, and platform-specific code can make AI integration expensive. A company cannot simply attach a model to badly organized data and expect clean answers. Garbage in, faster garbage out.
AI Is Becoming Part of Platform Security
Cybersecurity accounts for 8.3 percent of weighted AI activity in the report, while incident management and development operations account for another 5.3 percent. Fraud, payments, and anti-money-laundering systems add 5.6 percent. This part of the data has direct relevance to multiplayer gaming.
Competitive platforms already fight account theft, botting, smurfing, payment fraud, chat abuse, boosting, ban evasion, match manipulation, and cheat distribution. Machine-learning systems can analyze activity at a scale that human moderation teams cannot match. They can flag unusual aim patterns, impossible movement, suspicious login behavior, coordinated account networks, abnormal marketplace transactions, and repeated harassment across large message histories. The benefit is speed. The risk is false confidence.
A detection model is not automatically fair because it processes millions of records. It can inherit poor assumptions from training data, punish unusual but legitimate players, and create opaque enforcement decisions that support teams cannot explain. Players who receive an automated ban rarely care that the system was statistically accurate across the broader population. They care that their account is gone.
The strongest approach keeps AI in an investigative role. Models identify suspicious behavior, rank cases, connect evidence, and help human reviewers work faster. High-impact penalties still need clear standards, retained evidence, and an appeal process that reaches an actual person. That may cost more than fully automated enforcement. It also prevents an esports platform from turning its integrity system into an unaccountable black box.
Generative AI Is Common, Autonomous AI Is Not
The report draws a sharp line between generative tools and agentic systems. More than four out of five companies report using generative AI, but only 32.1 percent report activity involving AI agents or reasoning systems. By comparison, a broader KPMG enterprise survey found that 98.3 percent of respondents had some degree of engagement with AI agents, with roughly one-third already scaling them across several business functions.
Gaming’s slower movement makes sense. Autonomous systems are more dangerous in environments involving financial transactions, player protection, regulated outcomes, live services, or competitive integrity. A chatbot that drafts an internal memo can be corrected. An autonomous system that changes player offers, blocks transactions, adjusts game parameters, or takes enforcement action can produce immediate damage.
Video game publishers will face the same boundary. An AI assistant suggesting code is one thing. An agent changing live matchmaking rules, modifying item prices, deploying server updates, or tuning competitive balance without direct review is something else entirely.
Agentic tools may eventually manage build pipelines, monitor server health, generate routine support responses, reproduce bugs, and open development tickets. Full autonomy will remain rare anywhere a bad decision can affect millions of players at once.
Studios do not need artificial executives running the game overnight. They need dependable systems that can perform limited work, report what they changed, and stop when confidence falls below an established threshold.
The Cost-Saving Promise Has Not Been Proven
Cost reduction is the strongest stated motivation for AI adoption. In the survey, 42.2 percent selected reduced costs as their primary investment driver, while 33.7 percent selected revenue growth. Compliance and regulatory risk management came a distant third at 9.6 percent. The financial results are far less impressive.
Only one in five companies reports meaningful returns from AI. Most expect returns within two years, but one-quarter have no structured process for evaluating success. Average reported cost savings scored only 2.43 out of 5. Nearly half said AI had produced savings only to a small extent, while 10.8 percent reported none. Just 8.4 percent described the savings as large or transformative.
Revenue results are somewhat better. About 54.2 percent reported at least some positive revenue effect, but most of that group described the increase as slight. Only 10.8 percent reported a significant increase.
This is the number studio employees should remember during the next corporate presentation about AI efficiency. Adoption does not prove savings. A company can buy licenses, build internal tools, train employees, hire consultants, review generated output, settle legal questions, and add oversight staff without reducing its total production budget. AI can shift costs rather than remove them.
The pressure will still fall on headcount because payroll is visible and easy to cut. The report notes that nearly half of surveyed organizations have no AI-specific hiring planned, even though training gaps, limited expertise, and resource constraints are among their most frequently reported obstacles. That is not a technology plan. It is an accounting plan wearing a technology badge.
Governance Is the Weakest Part of the Industry
Only one in five surveyed companies has a dedicated AI governance role. Responsible AI practices exist in some form at about 70 percent of companies, but fewer than 5 percent describe those practices as fully embedded. The AiR Hub’s detailed findings place that fully embedded figure at only 2 percent. Nearly one-third report no established responsible-AI practices at all.
Regulators are not convinced that businesses can police themselves. Fifty-eight percent of regulatory respondents said the industry cannot effectively self-regulate its AI use. Only 13 percent knew of licensed operators with internal responsible-AI policies or frameworks.
Video game companies may not face casino regulators, but they do face platform policies, privacy laws, labor agreements, consumer-protection rules, intellectual-property disputes, and growing demands for disclosure. Studios using generated assets need records showing where tools were applied, which models were approved, how outputs were reviewed, and whether protected material entered the production pipeline.
Competitive platforms carry an even heavier burden. Any machine-learning system affecting matchmaking, moderation, rankings, player safety, monetization, or disciplinary action needs documented limits and a human appeal path.
The current data shows an industry racing to install AI in code, products, security, and customer operations while the systems for accountability remain unfinished. That gap will define the next stage of adoption, especially once players begin demanding proof that automated decisions are accurate, explainable, and reversible.